CN113677260B - System for identifying sleep disorders including sensing unit and data processing device - Google Patents

Abstract

The present invention relates to devices, systems and methods for detecting disturbances that may occur during sleep of a subject.

Description

System for identifying sleep disorders comprising a sensing unit and a data processing device

Technical Field

The present invention relates to devices, systems and methods for detecting disturbances that may occur during a subject's sleep.

Background Art

The most common method published for evaluating sleep disorders and more specifically sleep-disordered breathing is laboratory polysomnography (PSG). This test requires staying overnight in a dedicated facility supervised by a trained technician. However, this method is expensive, time-consuming and cannot keep up with demand. During the PSG test, multiple physiological signals are recorded by different types of sensors (e.g., EEG, EMG, ECG, thermistor, pressure, video). The data from these sensors are checked by health care professionals later.

Alternative systems are considered in the art, US 2017/0265801 relates to a bruxism detection system for detection of teeth grinding and tapping. The system includes an accelerometer mounted on the jaw that senses and records changes in acceleration at the beginning and end of jaw clenching. Data from the accelerometer is processed to distinguish movements associated with bruxism from other movements of the head by comparison with an accelerometer threshold.

US2017/0035350 also relates to a bruxism detection system. The system includes two accelerometers mounted on the masseter muscles, the first accelerometer is attached to the skin of the left masseter muscle, and the second accelerometer is attached to the skin of the right masseter muscle. When the recorded data of the two accelerometers are substantially equal, bruxism is detected.

US2007/273366 relates to a sleep disorder detector system. The system comprises a device for measuring distance by detecting an emitted magnetic field. The device may be mounted on a support arranged to be applied to the head to measure the movement of the mouth. Data from the device is processed to detect sleep breathing disorders, such as snoring.

A problem with the known system is that the movements of the head of the subject wearing the sensing unit and the movements of their mandible are considered separately from one another. The same applies to the position of the head and the position of the mandible, which are calculated based on the movements measured by the accelerometers. However, the data from the accelerometers are limited and may be influenced by the movements of other body parts, such as the chest or the trachea during breathing. Therefore, the connection between these different movements and positions cannot be fully taken into account for the analysis of sleep disturbances, which may have a negative impact on the diagnosis based on the measured data stream.

In fact, jaw movement can be induced by respiratory or non-respiratory movements. Therefore, the movement of the head when a person is sleeping may cause the movement of their mandible. The mandible can be considered as a mechanical connection to the tracheal traction or an effector controlled by the brain. Therefore, jaw movement can be passively induced by the respiratory movement of the tracheal traction, or directly controlled by the brain. The traction is the traction exerted by the chest on the head of a person. This traction has the breathing frequency of the person because the chest moves with the breathing, because the respiratory muscles are controlled by the brain. Therefore, if the head moves at the breathing frequency, the mandible attached to the head will follow the movement exerted by the head at the breathing frequency. This is a passive movement that follows the movement of the head. Even when the head may not move, or most often does not move, the jaw movement can also be directly and actively controlled by the brain. The brain controls the jaw movement by stimulating a set of jaw muscles. Therefore, it is useful to be able to distinguish between jaw movements controlled by the brain or by the attached tracheal traction. A system that can more accurately interpret signals from the brain and more accurately identify sleep disorders is needed.

Summary of the invention

It is an object of the present invention to provide a system comprising a sensing unit and a data processing device for correlating measurements of the movement and position of a subject's head and mandible in time during analysis of the measurement data.

In particular, the invention relates to a system (equivalently, a combination) comprising a sensing unit comprising a gyroscope adapted to measure the movement of the mandible of the subject and a processing unit for processing data related to disturbances that may occur during the sleep of a subject. The inventors surprisingly found that the use of a gyroscope allows the capture of the mandibular spin and therefore the assessment of the activity of the brainstem controlling the mandibular movement during sleep.

In some embodiments, the invention relates to a system comprising a sensing unit and a device (e.g., a processing unit) for processing data related to disturbances that may occur during the sleep of a subject, the sensing unit comprising an accelerometer adapted to measure movements of the subject's head and/or mandible and a gyroscope adapted to measure movements of the subject's mandible. The sensing unit is adapted to generate a measurement signal based on the measurements achieved, and the processing unit comprises a first input and a second input for receiving a first time stream of measurement signals from the accelerometer and a second time stream of measurement signals from the gyroscope, respectively.

Therefore, a system for characterizing sleep disorders in a subject having a head and a mandible is provided herein, the system comprising a gyroscope, a data analysis unit, the gyroscope and the data analysis unit being connected via a data link. In a specific embodiment, the system is characterized by comprising:

- a gyroscope configured to measure rotational movement of a mandible of the subject;

- a data analysis unit and a data link configured for sending the measured rotational movement data from the gyroscope to the data analysis unit;

wherein the data analysis unit comprises a memory unit configured to store N mandibular movement categories, wherein N is an integer greater than 1, wherein at least one of the N mandibular movement categories represents a sleep disorder event;

- wherein each j-th (1≤j≤N) mandibular motion category comprises a j-th rotation value set, each j-th rotation value set representing at least one rate, rate change, frequency and/or amplitude of mandibular rotation associated with the j-th category;

- wherein the data analysis unit comprises a sampling element configured to sample the measured rotational motion data during a sampling period, thereby obtaining sampled rotational motion data;

- wherein the data analysis unit is configured to derive a plurality of measured rotation values from the sampled rotational motion data; and

- wherein the data analysis unit is further configured for matching the measured rotation values to N mandibular movement categories.

In some embodiments, the system includes an accelerometer adapted to measure acceleration indicative of movement and/or position of the subject's head and/or mandible,

- the data link is further configured for transmitting the measured acceleration data from the accelerometer to the data analysis unit;

- wherein each j-th (1≤j≤N) mandibular motion category comprises a j-th set of acceleration values, each j-th set of acceleration values representing at least one mandibular motion or head motion associated with the j-th category;

- wherein the sampling element is configured to sample the measured acceleration data during a sampling period, thereby obtaining sampled acceleration data;

- wherein the data analysis unit is configured to derive a plurality of measured acceleration values from the sampled acceleration data; and

- wherein the data analysis unit is further configured for matching the measured acceleration values to N mandibular movement categories.

In some embodiments, the system further comprises a magnetometer adapted to measure magnetic field data, changes in the magnetic field data being indicative of movement of the subject's head and/or mandible,

- the data link is further configured for sending the measured magnetic field data from the accelerometer to the data analysis unit;

- wherein each j-th (1≤j≤N) mandibular motion category comprises a j-th set of magnetic field data values, each j-th set of magnetic field data values representing at least one rate or rate change of mandibular motion or head motion associated with the j-th category;

- wherein the data analysis unit comprises a sampling element configured to sample the measured magnetic field data during a sampling period, thereby obtaining sampled magnetic field data;

- wherein the data analysis unit is configured to derive a plurality of measured magnetic field values from the sampled magnetic field data; and

- wherein the data analysis unit is further configured to match the measured magnetic field values to N mandibular movement categories.

In some embodiments, the gyroscope and optionally the accelerometer and/or the magnetometer or a portion of the gyroscope and optionally the accelerometer and/or the magnetometer is included in a sensing unit, which is mountable on the mandible of the subject.

In some embodiments, one or more of the N mandibular motion categories are characterized by a predetermined frequency range.

In some embodiments, the analysis unit is configured for identifying movement of the subject's head based on the gyroscope data, based on the accelerometer data and/or the magnetometer data.

In some embodiments, at least one of the N mandibular movement categories indicates that the subject is awake, and multiple of the N mandibular movement categories indicate that the subject is asleep.

In some embodiments, at least one of the N mandibular movement categories indicates that the subject is in an N1 sleep state; wherein at least one of the N mandibular movement categories indicates that the subject is in a REM sleep state; optionally, wherein at least one of the N mandibular movement categories indicates that the subject is in an N2 sleep state, and/or wherein at least one of the N mandibular movement categories indicates that the subject is in an N3 sleep state.

In some embodiments, one or more of the N mandibular movement categories represent obstructive apnea, obstructive hypopnea, respiratory effort associated with arousals, central apnea, and/or central hypopnea.

In some embodiments, one of the N mandibular motion categories represents grinding, wherein the measured rotational motion data represents a mandibular motion amplitude of at least 1 mm at a frequency established in the range of 0.5 Hz to 5 Hz during at least three respiratory cycles when the movement is staged, or a mandibular motion amplitude exceeding 1 mm for at least 2 seconds in a sustained, tense manner.

Also provided is a method for assisting in characterizing a sleep disorder in a subject having a mandibular sac, the method comprising the steps of:

- receiving, by the data analysis unit and via a data link, rotational movement data from a gyroscope located at the mandible of the subject;

- storing, by means of a memory unit comprised in the data analysis unit, N mandibular movement categories, wherein N is an integer greater than 1, wherein at least one of the N mandibular movement categories represents a sleep disorder event;

- wherein each j-th (1≤j≤N) mandibular motion category comprises a j-th rotation value set, each j-th rotation value set representing at least one rate, rate change, frequency or amplitude of mandibular rotation associated with the j-th category;

- sampling the rotational motion data during a sampling period by means of a sampling element included in the data analysis unit, thereby obtaining sampled rotational motion data;

- deriving a plurality of measured rotation values from the sampled rotational motion data by means of a data analysis unit; and

- The measured rotation values are matched to the N mandibular movement classes by means of a data analysis unit.

In some embodiments, the method further comprises the following steps:

- measuring accelerations by means of accelerometers, the accelerations being indicative of the movement and/or position of the subject's head and/or mandible;

- sending the measured acceleration data from the accelerometer to a data analysis unit by means of a data link;

- wherein each j-th (1≤j≤N) mandibular motion category comprises a j-th set of acceleration values, each j-th set of acceleration values representing at least one mandibular motion or head motion associated with the j-th category;

- sampling the measured acceleration data during a sampling period by means of a sampling element, thereby obtaining sampled acceleration data;

- deriving a plurality of measured acceleration values from the sampled acceleration data by means of a data analysis unit; and

- The measured acceleration values are matched to the N mandibular movement classes by means of a data analysis unit.

In some embodiments, the method further comprises the following steps:

- measuring magnetic field data by means of a magnetometer, changes in the magnetic field data being indicative of movement of the subject's head and/or mandible;

- sending the measured magnetic field data from the magnetometer to a data analysis unit by means of a data link;

- wherein each j-th (1≤j≤N) mandibular motion category comprises a j-th set of magnetic field data values, each j-th set of magnetic field data values representing at least one rate or rate change of mandibular motion or head motion associated with the j-th category;

- sampling the measured magnetic field data during a sampling period by means of a sampling element included in the data analysis unit, thereby obtaining sampled magnetic field data;

- deriving a plurality of measured magnetic field values from the sampled magnetic field data by means of a data analysis unit; and

- The measured magnetic field values are matched to the N mandibular movement classes by means of a data analysis unit.

In some embodiments, the method further comprises the step of identifying, by means of the analysis unit, a movement of the subject's head based on the gyroscope data, based on the accelerometer data and/or on the magnetometer data.

In some embodiments, at least one of the N mandibular motion categories represents molars, wherein the measured rotational motion data represents a mandibular motion amplitude of at least 1 mm at a frequency established in the range of 0.5 Hz to 5 Hz during at least three respiratory cycles when the movement is staged, or a mandibular motion amplitude exceeding 1 mm for at least 2 seconds in a sustained, tense manner.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention will now be described in more detail with the aid of the accompanying drawings, which illustrate the system and its operation. The present system may be described as a system comprising a sensing unit and a device or unit for processing the sensed data. In the accompanying drawings:

Fig. 1 shows a system according to the invention.

2A and 2B show two flows during a change in position of the head of a person lying in a bed.

3A and 3B show the flow captured by the sensing unit during teeth grinding.

Figure 4 shows the loop gain.

FIG5 shows the identification of micro-arousals after pre-processing.

FIG. 6 shows the measured signal after applying a bandpass filter.

FIG. 7 shows signals representing micro-arousals.

FIG8 shows an example of a first time stream and a second time stream in the case of obstructive apnea;

FIG9 shows an example of a first time stream and a second time stream in the case of obstructive respiratory attenuation;

FIG10 shows an example of a first time stream and a second time stream in the case of mixed apnea;

FIG11 shows an example of a first time stream and a second time stream in the case of central apnea;

FIG12 shows an example of a first time stream and a second time stream in the case of central respiratory depression;

FIG13 shows an example of a first time stream and a third time stream during respiratory effort related to arousal (RERA); and

FIG. 14 shows a frequency spectrum diagram of the frequency distribution of jaw movement.

FIG15 illustrates an exemplary process for feature extraction, data processing, and data description.

16 and 17 illustrate the analysis of jaw movement data captured with the aid of a magnetic sensor.

FIG18 shows an exemplary method for automatic sleep stage detection based on jaw movement data captured with the aid of a gyroscope and an accelerometer. This method is further discussed in Example 18.

In FIG1 , the following reference numbers are used: 1 - sensing unit; 2 - accelerometer; 3 - gyroscope; 4 - magnetometer; 5 - oximeter; 6 - thermometer; 7 - audio sensor; 8 - electromyography unit; 9 - pulse photoplethysmograph; 10 - device for processing data; 11 - 1 - first input; 11 - 2 - second input; 11 - 3 - third input; 11 - 4 - fourth input; 12 - identification unit; 13 - analysis unit.

DETAILED DESCRIPTION

Before describing the systems and processes of the present invention, it should be understood that this is not limited to the specific systems and methods or combinations described, as such systems and methods and combinations may of course vary. It should also be understood that the terminology used herein is not intended to be limiting, as the scope will be limited only by the appended claims.

As used herein, the singular forms "a," "an," and "the" include singular and plural referents unless the context clearly dictates otherwise.

As used herein, the terms "comprising," "including," and "consisting of are synonymous with "encompassing," "comprising," or "containing," "comprising," and are inclusive or open-ended and do not exclude additional, unrecited members, elements, or method steps. It should be understood that the terms "comprising," "including," and "consisting of" as used herein include the terms "comprising," "consisting of," and "consisting of."

The recitation of numerical ranges by endpoints includes all numbers and fractions subsumed within the corresponding ranges, as well as the recited endpoints.

As used herein, the terms "about" or "approximately" when referring to a measurable value such as a parameter, amount, duration, etc., are intended to include variations of +/-10% or less of the specified value, preferably variations of +/-5% or less of the specified value, more preferably variations of +/-1% or less of the specified value, and still more preferably variations of +/-0.1% or less of the specified value, as long as such variations are suitable for implementation in the disclosed aspects and embodiments. It should be understood that the value referred to by the modifier "about" or "approximately" itself is also specifically and preferably disclosed.

Although the term "one or more" or "at least one", such as one or more or at least one member in a group of members, is clear in itself, by way of further illustration, the term specifically includes reference to any one of the members, or reference to any two or more of the members, for example, any ≥3, ≥4, ≥5, ≥6 or ≥7 of the members, etc., and up to all members.

All references cited in this specification are hereby incorporated by reference in their entirety. In particular, the teachings of all references specifically mentioned herein are hereby incorporated by reference.

Unless otherwise defined, all terms used herein, including technical and scientific terms, have the meanings as commonly understood by one of ordinary skill in the art. By way of further guidance, term definitions are included to better understand the teachings as described herein.

In the following paragraphs, different aspects are defined in more detail. Unless explicitly stated otherwise, each aspect so defined can be combined with any other one or more aspects. Specifically, any feature indicated as preferred, particular or advantageous can be combined with any other one or more features indicated as preferred, particular or advantageous.

The present invention relates to the measurement and assessment of jaw movements in sleeping subjects. The mandible or lower jaw bone is located below the maxilla and forms the chin. (Not taking into account the middle ear ossicles) it is the only movable bone of the human skull. During movement, the mandible pivots around the temporomandibular joint, where the mandible is attached to the skull (temporal bone) in front of the ear. During jaw movement, the relationship between the length and tension of the muscle fibers fixed to the mandible will change, which may lead to stiffening of the upper airway in subjects at risk of instability during sleep. This movement is activated under the agonist and antagonist muscles, which serve to raise or lower the mandible and thus close or open the mouth accordingly. The agonist and antagonist muscles are innervated by motor neurons originating from the nucleus of the trigeminal nerve located in the brainstem (mid-pons) and are supported by the motor branches of this nerve.

A system for characterizing sleep disorders in a subject having a head and mandible is provided herein. The system includes a gyroscope. The gyroscope is configured to measure the rotational motion of the subject's mandible, which, as observed by the inventors, is an activity for which the gyroscope is particularly suitable. The gyroscope can be used to assess the activity of the brainstem stimulating mandibular motion during sleep to keep the upper airway (pharynx) open and prevent sleep-disordered breathing. The mandibular movable bone rotates like a lever to stretch the pharyngeal muscle fibers (including the tongue) that are directly or indirectly attached to the mandibular arch (via the second movable bone-hyoid bone).

To the extent that gyroscopic movement is representative of central drive, it means that the nucleus of the trigeminal nerve in the pons is acting to subtly move the mandible relative to respiratory centers also located in the brainstem and under the influence of higher centers responsible for sleep organization (sleep staging). Therefore, providing a gyroscope in the sensing unit can be used to assess various sleep-related activities by looking at rotational mandibular displacement, which sleep-related activities can include breathing, sleep stages, or other events (e.g., movement or motor events). In addition, values measured by a gyroscope arranged to measure rotational movement of a subject's mandible (such as the rate and amplitude of the mandibular gyroscope signal) can be used to obtain an assessment of central drive originating from the nucleus of the trigeminal nerve, in addition to metrics derived directly or indirectly from the measurements.

The inventors have found that other sensing units are not suitable for the measurement and assessment of mandibular movements as provided herein. For example, inertial sensors like accelerometers only allow limited measurement of linear acceleration and are therefore not suitable for measuring rotational mandibular displacement. Accelerometer measurements may be affected by movements of the body or head (such as the chest or trachea during breathing), and distinguishing between data sources is difficult and adds unnecessary noise and complexity to the system. Therefore, existing systems for analyzing sleep disturbances do not fully consider the connection between possible body and head movements. This has a negative impact on the diagnosis based on the measured data stream. The inventors have found that the rotation of the mandible carries the information required for an accurate assessment and that this movement can moreover be accurately recorded by a gyroscope.

The system also includes a data analysis unit and a data link. The data link provides a communication path between the gyroscope and the data analysis unit. Preferably, the data link is a wireless data link, for example due to improved subject comfort, although a data link employing wired communication is of course possible.

The rotational motion data is sent from the gyroscope to a data analysis unit via a data link. The data link is of a conventional nature and comprises an arrangement for transmitting data wirelessly or by wire.

The data analysis unit includes a memory unit, for example, a data storage device such as a hard disk drive, a solid-state drive, a memory card, etc. The memory unit is configured to store a plurality of (N) mandibular motion specific patterns (categories), where N is an integer greater than 1. At least one of the N mandibular motion categories represents a sleep disorder event. Preferably, the N mandibular motion categories include a plurality of motion categories representing various mandibular motions. Each j-th (1≤j≤N) mandibular motion category includes a j-th rotation value set, and each j-th rotation value set represents at least one rate, rate change, frequency and/or amplitude of mandibular rotation associated with the j-th category.

The rotational motion data measured or recorded by the gyroscope are correlated with the mandibular motion categories as follows:

The data analysis unit comprises a sampling element configured to sample the measured rotational motion data during a sampling period. Thereby, sampled rotational motion data are obtained. Thus, information contained in the signal recorded by the gyroscope can be extracted for further analysis. It should be understood that in some embodiments, the data analysis unit can be included in a general-purpose computing device such as a personal computer or a smart phone, but it is of course possible to provide dedicated hardware.

The data analysis unit is configured to derive a plurality of measured rotation values from the sampled rotational motion data, and to match the measured rotation values with N mandibular motion categories. Preferably, deriving the measured rotation values from the sampled rotational motion data includes one or more of the following processes: discretization, time averaging, fast Fourier transform, etc. In addition, the matching can be fully or partially automated by providing a machine learning model, so that the data analysis unit is configured to learn a plurality of statistical and/or physical metrics in order to capture the features of the signal in the frequency domain and the time domain and identify patterns of the rotation signal for specific events (such as sleep stages, breathing efforts, etc.). Therefore, the provision of a machine learning model can provide automatic interpretation of relevant information and/or match feature data with sleep disorder events.

Therefore, the study of jaw movements during sleep provides information about the state of respiratory control in response to changes in permeability or resistance to the flow of air in the upper airway, whether or not it involves a series of modifications in head position.

Analysis of the nature of the jaw movements using the system according to the invention also allows detection of repetitive non-respiratory movement events during sleep, such as teeth grinding or chewing, or isolated such as orofacial movement disorders. Swallowing and sucking movements in infants can also be clearly identified. In addition, swallowing movements can also be detected in adults. This allows distinguishing arousals from micro-arousals.

In some embodiments, one or more jaw movement categories represent isolated large jaw movements (IMMs). IMMs are associated with micro-arousals or arousals induced by respiratory disturbances, so that micro-arousals can be effectively inferred from measurement and analysis.

In some embodiments, the process of matching the measured rotation values to the N mandibular motion categories utilizes an artificial intelligence approach, such as a random forest.

In some embodiments, the system further comprises an accelerometer. The accelerometer is adapted to measure acceleration (including changes in acceleration) which is representative of movement and/or position of the subject's head and/or mandible. The inventors have found that accelerometers are particularly suitable for measuring movement and position of the head. Adding an accelerometer to the present system allows further assessment of the behavior of the mandible during sleep. In particular, the inventors have found that measurements of acceleration can be used to account for unexpected changes in the movement, amplitude and/or rate of the gyroscope. Thus, measurements of the accelerometer can be used to supplement the measurements performed by the gyroscope.

The measured or recorded acceleration data is sent by the accelerometer to the data analysis unit via a data link. In these embodiments, each j-th (1≤j≤N) mandibular motion category includes a j-th set of acceleration values. Each j-th set of acceleration values or metric represents at least one mandibular motion or head motion associated with the j-th category. The sampling element is configured to sample the measured acceleration data during a sampling period. After sampling, the measured acceleration data is referred to as sampled acceleration data. The data analysis unit is configured to derive a plurality of measured acceleration values from the sampled acceleration data, for example by discretization and optionally time averaging. Information contained in the signal recorded by the accelerometer can be extracted for further analysis. The data analysis unit is also configured to match the measured acceleration values with N mandibular motion categories. This matching process is understood to involve automatically determining the mandibular motion category that is closest to the measured acceleration value. Matching can be fully or partially automated by providing a machine learning model that can provide automatic interpretation of relevant information and/or match feature data with sleep disorder events.

The inventors have discovered that accelerometers are particularly sensitive to movement of the head. Together, a gyroscope and an accelerometer allow head movement to be effectively distinguished from mandibular movement, which in turn allows for improved detection of sleep disorder events. Therefore, providing a gyroscope and an accelerometer in a single system can increase the sensitivity and accuracy of the system, and can also be used to evaluate new information that cannot be explained from measurements provided by a gyroscope or accelerometer alone. For example, changes in head position stimulated by central activation may affect mandibular rotational movement, which may be incorrectly interpreted as changes in mouth opening or mouth closure. Therefore, the combination of a gyroscope and an accelerometer can allow head movement to be distinguished from jaw movement. In view of the superior and unexpected functionality provided by the present combination, the presence of a gyroscope cannot be considered a substitute for other sensing devices (such as, for example, a second accelerometer).

In some embodiments, the system further comprises a magnetometer adapted to measure magnetic field data. Changes in the magnetic field data represent the direction and/or position of movement of the subject's head and/or mandible. Adding a magnetometer to the present system may allow further assessment of the behavior of the mandible during sleep. It will be appreciated that a magnetometer is provided in the present system for assessing the orientation of a compass-like sensing unit. Therefore, the magnetometer is not intended to be used as a unit for measuring distance as contemplated in systems of the art; although the primary function is understood to not limit the scope of the present system.

The data link is also configured to send the measured or recorded magnetic field data from the magnetometer to the data analysis unit. Each j-th (1≤j≤N) mandibular motion category includes a j-th magnetic field data value set. Each j-th magnetic field data value set represents at least one rate or rate change of mandibular motion or head motion associated with the j-th category. The data analysis unit includes a sampling element, which is configured to sample the measured magnetic field data during a sampling period. The sampled magnetic field data is thereby obtained. The data analysis unit is configured to derive a plurality of measured magnetic field values from the sampled magnetic field data. The data analysis unit is also configured to match the measured magnetic field values with N mandibular motion categories.

In a particular form, the magnetometer may comprise two parts: one part mounted on the patient's forehead and one part mounted on the patient's mandible. The inventors have found that this is a particularly effective configuration for detecting jaw movement.

In some embodiments, signals originating from magnetometers, gyroscopes, accelerometers, and/or other sensors are transmitted over a single physical medium using, for example, time division multiplexing and/or using carrier waves of different frequencies.

In some embodiments, a gyroscope, and/or an accelerometer, and/or a magnetometer, or a portion thereof, is included in the sensing unit. The sensing unit may be mounted on the mandible of the subject. This is an embodiment with a highly compact form factor, which is easy to apply and provides improved patient comfort. The provision of an accelerometer and/or a magnetometer is understood not to replace the functionality of the gyroscope, but to obtain a new interpretation that is only possible through the combination of a gyroscope with one or more additional sensing devices (such as an accelerometer and/or a magnetometer). Preferably, first, the interpretation of the acquisition signal is associated with the data from the gyroscope, and in a second step is supplemented with the data from the accelerometer and/or the magnetometer. For example, first, the data from the gyroscope can be used to analyze the angular velocity of the mandible to obtain a comprehensive cycle by cycle analysis; then, the data from the accelerometer can be used to provide the background of the generation of the cycle (e.g., the origin of the activation (cortical and subcortical), the endotype (dynamic) of the (disturbed breathing), the type of (more or less tense or phasic) muscle chewing activity). Additionally, new assessments can be made based on combinations of data that would not be possible based on data from a single sensing unit alone. For example, a precise description of the event type opens up the possibility of making predictions about the occurrence or recurrence of sleep disorder events or respiratory changes (e.g., peripheral capillary oxygen saturation _SpO2 ).

Preferably, the sensing unit has dimensions of at most 5 cm long, 2 cm thick and 1 cm high. This reduces disruption to the subject's normal sleep.

In some embodiments, one or more of the N mandibular motion categories are associated with a predetermined frequency range. In other words, in these embodiments, one or more of the N mandibular motion categories include mandibular motion occurring within a predetermined frequency range. Preferably, at least two of the N mandibular motion categories are associated with a predetermined frequency range, the predetermined frequency ranges include an Ath predetermined frequency range and a Bth predetermined frequency range, and the Ath predetermined frequency range and the Bth predetermined frequency range do not overlap.

In some embodiments, the at least one predetermined frequency range includes frequencies between 0.15 Hz and 0.60 Hz, or frequencies between 0.25 Hz and 0.50 Hz, or frequencies between 0.30 Hz and 0.40 Hz. This is the frequency range of the signal representing the subject's breathing.

In some embodiments, the system further comprises one or more auxiliary components selected from the list comprising an oximeter and/or a thermometer and/or an audio sensor and/or an electromyography unit and/or a pulsed photoplethysmograph. Preferably, these auxiliary components are operatively connected to the analysis unit via a data link.

In some embodiments, the analysis unit is configured to identify the movement of the subject's head based on the gyroscope data, and/or the accelerometer data, and/or the magnetometer data. Preferably, the movement of the head comprises a rotation, such as a rotation around an axis passing through the center of the subject's head. Then, preferably, at least one of the N mandibular motion categories represents a change in the position of the head. This allows general head movement to be effectively distinguished from the movement of the mandible itself. In these embodiments, the system preferably comprises both an accelerometer and a gyroscope.

In some embodiments, the analysis unit is adapted to apply one or more pre-processing steps to the gyroscope data, and/or the accelerometer data, and/or the magnetometer data. The one or more pre-processing steps are selected from the list comprising: application of a band pass filter, application of a low pass filter, exponential activity average, and/or calculation of entropy of the frequency of the gyroscope data and/or the accelerometer data and/or the magnetometer data. The application of low pass filtering improves the detection of micro-arousals.

In some embodiments, the analysis unit may include an interpretation module that is configured to interpret specific parameters that measure the extension of sleep quality and sleep breathing disturbances. Sleep quality parameters may include, for example, total sleep time (TST), sleep onset latency (SOL), first awakening from sleep (WASO), wakefulness index, sleep efficiency (SE), the ratio of REM, non-REM sleep, REM sleep latency, and other sleep quality metrics. Metrics related to sleep breathing disturbances may include the rate of occurrence per hour during sleep and the cumulative duration of respiratory effort. The analysis unit may be configured to report the subject-specific parameters explained. The report may include providing the output to a device such as a computer or smart phone. The report may also include, for example, providing a visual or textual report of subject-specific parameters in the form of a sleep graph.

In some embodiments, at least one of the N mandibular movement categories indicates that the subject is awake, wherein multiple of the N mandibular movement categories indicate that the subject is asleep. Incorporating the classifications of "asleep" and "awake" in the present method ensures that measurements taken when the subject is awake or asleep are interpreted accordingly. The interpretation module can be used to perform the interpretation.

In some embodiments, at least one of the N mandibular movement categories indicates that the subject is in an N1 sleep state; and at least one of the N mandibular movement categories indicates that the subject is in a REM sleep state. Optionally, at least one of the N mandibular movement categories indicates that the subject is in an N2 sleep state, and/or at least one of the N mandibular movement categories indicates that the subject is in an N3 sleep state.

In some embodiments, at least one of the N mandibular movement categories indicates that the subject is in an N2 sleep state.

In some embodiments, at least one of the N mandibular movement categories indicates that the subject is in an N3 sleep state.

In some embodiments, one or more of the N mandibular motion categories are associated with detection of sleep stages. Detection of sleep stages may further be implemented to establish subject specific sleep patterns. Sleep stage detection is preferably performed automatically at different resolution levels.

In a preferred embodiment, sleep modes (classified by increasing level of complexity) may include:

(1) 2-category (i.e., binary) scoring to detect the subject’s wakefulness or sleep state;

(2) a 3-category score for classifying sleep stages including wakefulness, non-REM sleep stages, or REM sleep stages of the subject;

(3) a 4-category score for classifying the subject's sleep stage into wakefulness, light sleep (N1 and N2) stages, deep sleep (N3) stage, or REM sleep stage;

(4) A 5-category score for classifying all sleep stages including the subject's wakefulness, N1 sleep stage, N2 sleep stage, N3 sleep stage, and REM sleep stage.

Exemplary methods for implementing automatic sleep stage detection with 3-category scoring are provided in Examples 18 and 19.

In some embodiments, at least one of the N mandibular motion categories represents cortical activity.

In some embodiments, at least one of the N mandibular motion categories represents subcortical activity.

In some embodiments, one or more of the N mandibular movement categories represent obstructive apnea, obstructive hypopnea, respiratory effort associated with arousals, central apnea, and/or central hypopnea.

In some embodiments, one of the N mandibular motion categories represents grinding and the measured rotational motion data represents a mandibular motion amplitude of at least 1 mm, a mandibular motion amplitude at a frequency established in the range of 0.5 Hz to 5 Hz during at least three respiratory cycles when the movement is staged, or a mandibular motion amplitude exceeding 1 mm for at least 2 seconds in a sustained, tense manner.

Between 5% and 10% of adults frequently complain of teeth grinding during sleep. The grinding is usually intermittent and variable over time, sometimes disappearing for a few weeks before rebounding and then re-imposing during the night for several consecutive nights. Grinding is often recognized by the sleeper's partner in the form of inconsistent and loud grinding of the teeth. It may cause facial or temporal pain in the subject and signs of tooth enamel wear. Its origin is unknown, but the syndrome of obstructive sleep apnea has been suggested as a possible cause.

In some embodiments, one or more of the N mandibular movement categories represent loop gain, muscle gain to mobilize the mandible during apnea or hypopnea or a period of effort, passive collapsibility point after activation, and/or arousability point before activation.

Further provided herein is a method for assisting in characterizing a sleep disorder (e.g., sleep-disordered breathing (SDB)) in a subject having a mandibular sac. The method comprises the following steps:

- Receiving rotational movement data from a gyroscope located at the mandible of the subject by a data analysis unit and via a data link.

- With the aid of a memory unit included in the data analysis unit, N mandibular motion categories are stored. It should be noted that N is an integer greater than 1, and at least one of the N mandibular motion categories represents a sleep disorder event (e.g., a sleep disordered breathing (SDB) event). Each j-th (1≤j≤N) mandibular motion category includes a j-th rotation value set, and each j-th rotation value set represents at least one rate, rate change, frequency, or amplitude of mandibular rotation associated with the j-th category.

- Sampling the rotational motion data during a sampling period by means of a sampling element comprised in the data analysis unit. Sampled rotational motion data are thereby obtained.

- deriving a plurality of measured rotation values from the sampled rotational motion data by means of a data analysis unit; and

- The measured rotation values are matched to the N mandibular movement classes by means of a data analysis unit.

Sleep disorders can therefore be detected effectively with excellent patient comfort.

In some embodiments, the method further comprises the following steps:

- measuring acceleration with the aid of an accelerometer. Acceleration indicates the movement and/or position of the subject's head and/or mandible;

- sending the measured acceleration data from the accelerometer to a data analysis unit by means of a data link;

- sampling the measured acceleration data during a sampling period by means of a sampling element, thereby obtaining sampled acceleration data;

- deriving a plurality of measured acceleration values from the sampled acceleration data by means of a data analysis unit;

- matching the measured acceleration values with the N mandibular movement categories by means of the data analysis unit. It should be noted that in these embodiments, each j-th (1≤j≤N) mandibular movement category comprises a j-th set of acceleration values, each j-th set of acceleration values representing at least one mandibular movement or head movement associated with the j-th category.

The use of both an accelerometer and a gyroscope allows for effective discrimination of mandibular motion from whole head motion.

In some embodiments, the method further comprises the following steps:

- measuring magnetic field data by means of a magnetometer, changes in the magnetic field data being indicative of movement and/or position of the subject's head and/or mandible;

- sending the measured magnetic field data from the accelerometer to a data analysis unit by means of a data link;

- sampling the measured magnetic field data during a sampling period by means of a sampling element included in the data analysis unit, thereby obtaining sampled magnetic field data;

- deriving a plurality of measured magnetic field values from the sampled magnetic field data by means of a data analysis unit; and

- matching the measured magnetic field values with the N mandibular movement categories by means of a data analysis unit. It should be noted that in these embodiments, each j-th (1≤j≤N) mandibular movement category comprises a j-th set of magnetic field data values, each j-th set of magnetic field data values representing at least one rate or rate change of mandibular movement or head movement associated with the j-th category.

In some embodiments, the method further comprises the step of identifying, by means of the analysis unit, a movement of the subject's head based on the gyroscope data, and/or based on the accelerometer data, and/or on the magnetometer data.

In some embodiments, at least one of the N mandibular motion categories represents grinding, and the measured rotational motion data represents a mandibular motion amplitude of at least 1 mm, a mandibular motion amplitude at a frequency established in the range of 0.5 Hz to 5 Hz during at least three respiratory cycles when the motion is staged, or a mandibular motion amplitude exceeding 1 mm for at least 2 seconds in a sustained, tense manner. This combination of parameters represents grinding, so that grinding can be effectively detected. In some embodiments, this frequency range is between 1.0 Hz to 4.5 Hz, or 1.5 Hz to 4.0 Hz, or 2.0 Hz to 3.5 Hz, or 2.5 Hz to 3.0 Hz.

In the following, specific embodiments of matching data (e.g., preferably sampled rotation, acceleration data and/or magnetic field data) with N mandibular motion categories are discussed. These embodiments involve extracting features from the above data. Features include measured rotation values, and optionally measured acceleration values and/or measured magnetic field values. Once the features are extracted, the features are matched with one or more mandibular motion categories. Preferably, the mandibular motion categories matched with the features include central hypopnea, normal sleep, and obstructive hypopnea. Preferably, the features are matched with the mandibular motion categories with the aid of SHAP scores to illustrate and explain the match.

In some embodiments, the features are selected from a non-exhaustive list including: central tendency (mean, median and mode) of MM (i.e., jaw movement measured using gyroscopes, accelerometers and/or magnetometers, representing rotation, acceleration and/or position) amplitude; MM distribution (raw or envelope signal): skewness, kurtosis, IQR, 25th percentile, 75th percentile and 90th percentile; extreme values: minimum, maximum, 5th percentile and 95th percentile of MM amplitude; trend of change: coefficients of linear trend and spline factors (S1, 2, 3, 4) based on tensor product from generalized additive model to evaluate MM in time function; duration of each event. It should be understood that such features refer to the measured rotation values, measured acceleration values and/or magnetic values, whether sampled and/or discretized. Preferably, the above values are sampled and discretized. It should be understood that this list presents exemplary embodiments, and thus, the exemplary embodiments are to be considered non-limiting of the present system.

In some embodiments, the extraction of features includes isolating events. An event is a series of mandibular motion data (preferably sampled rotation, acceleration and/or magnetic data) that can be attributed to a single movement of the head and/or mandible. A specific type of event is normal breathing, such as normal breathing for a predetermined amount of time. The predetermined amount of time can be, for example, between 2 seconds and 20 seconds, or between 5 seconds and 15 seconds, or between 5 seconds and 30 seconds, or between 5 seconds and 10 seconds. The time range size can be adapted to the intended application; for example, 30 seconds may be suitable for identifying sleep stages, 10 seconds for sleep bruxism or micro-awakenings, 20 seconds for respiratory events, etc.

In some embodiments, the feature extraction follows the following process, which includes steps 1 to 4:

1. Obtain sampled mandibular motion data. The mandibular motion data comprises sampled rotation values, and optionally sampled acceleration values and/or sampled magnetic field values. Preferably, the sampling rate is 1.0 Hz to 100.0 Hz, or 2.0 Hz to 50.0 Hz, or 5.0 Hz to 25.0 Hz, preferably 10.0 Hz. Preferably, the obtained sampled mandibular motion data is obtained during a time period between 10.0 minutes and 12.0 hours, or during a time period between 20.0 minutes and 4.0 hours, or during a time period between 30.0 minutes and 2.0 hours.

2. Timestamp the mandibular movement events.

3. For each timestamp ti, perform the following steps:

3.a. Check whether ti is the start of the mandibular movement event;

3.b. If ti is the start of the mandibular movement event,

- assigning ti to t_begin and then searching for the end of the mandibular movement event (t_end); and

- indexes t_begin and t_end;

4. For each mandibular movement event E, perform the following steps:

4.a. Calculate the event duration dt = (t_end-t_begin)

4.b. Determine the statistical distribution of the jaw movement data sampled during the event. Preferably, this involves calculating one or more features selected from the following list, the list comprising: minimum, maximum, mean, median, mode, 5th percentile, 25th percentile, 75th percentile, 90th percentile, 95th percentile, skewness, kurtosis, IQR;

Additionally or alternatively, a GAM (Generalized Additive Model) nonlinear model is used to estimate the MM amplitude and/or position by a spline function over time t, and then the coefficients of the spline function are extracted.

Additionally or alternatively, a simple linear model is fitted and the intercept and slope extracted from the jaw motion (including amplitude and/or position).

Optionally, concatenate all features.

Then, the mandibular movement events were matched with the mandibular movement categories.

In some embodiments, matching mandibular movement events to mandibular movement categories involves using exploratory data visualization, one-way ANOVA, and paired student-t tests with Bonferroni correction. Preferably, during this process, the significance level is set to p=0.0001 to 0.01, more preferably to p=0.001.

In some embodiments, machine learning methods (e.g., extreme gradient boosting, deep neural networks, convolutional neural networks, random forests) are used to classify the measured mandibular motion data into mandibular motion categories.

In some embodiments, the random forest method algorithm employed employs between 20 and 5000, or between 100 and 2000, or between 200 and 1000, or 500 decision trees. In some embodiments, each decision tree is constructed on a random subset of the above features.

In some embodiments, model development (i.e., training an artificial intelligence method) involves randomly splitting the measured mandibular motion data into two subsets, with the larger set used for model development and the smaller set used for model validation. In some embodiments, the larger set includes 60% to 80%, or 70% of the measured mandibular motion data. In some embodiments, the smaller set includes 20% to 40% or 30% of the mandibular motion data. Preferably, a synthetic minority over-sampling technique (SMOTE) is used on the training set before developing the model.

In some embodiments, model development involves evaluating the contribution of multiple features to classification with the aid of Lundberg's Shapley additive explanation (SHAP) method. Thus, the SHAP method allows for explanation of predictions made by the employed machine learning model; it allows the model to be interpretable.

Certain aspects of the disclosure may alternatively or additionally be expressed as follows:

In some embodiments, the system comprises a sensing unit and a device for processing data related to disturbances that may occur during the sleep of a subject. The processing device comprises an identification unit adapted to identify a first signal and a second signal in a first time stream and a second time stream, the frequency of the first signal being within a first predetermined frequency range, the value of at least one intrinsic characteristic characterizing the movement of the head and/or mandible in the second signal being within a second predetermined range consisting of a plurality of values; the first predetermined frequency range and the second predetermined range being respectively composed of the value of the frequency and the value of the movement of the head and mandible of the subject characterizing the sleep state of the subject; the identification unit being adapted to generate a trigger signal after observing the presence of the first signal and the second signal that have been identified in the first time stream and the second time stream for a first predetermined period of time; the identification unit being further adapted to identify a third signal in the first time stream and the second time stream after it has generated the trigger signal, the frequency and/or value of the at least one intrinsic characteristic in the third signal representing the movement of the mandible and/or the change in the position of the head of the subject; the identification unit being connected to an analysis unit adapted to be activated under the control of the trigger signal; the analysis unit being further adapted to compare the third signal with a profile characterizing the frequency and/or value associated with sleep disturbances and to generate a result of the comparison. The invention is based on the concept that during a subject's sleep, the subject's respiratory movements are controlled by nerve centers of the subject's brain, which control the muscles of the head and mandible to which they are attached, which will then position the subject's head and mandible. The accelerometers as well as the gyroscopes will each provide a corresponding temporal stream of measurement signals, which characterize the movements of the head and mandible. The use of an identification unit makes it possible to identify among these measurement signal streams those measurement signals that characterize the subject's sleep state, and thus activate the analysis unit to analyze any sleep disturbances affecting the subject when the subject is actually asleep.

It has thus been found that the movement of the mandible is determined not only by the movement of the chest but directly by the nerve centers of the brain which control the muscles attached to it and which will position the mandible. The nerve centers of the brain also control the position of the head.

In fact, the tracheal traction, which is necessarily at the respiratory rate, may cause head movements, and it is for this reason that measurements performed both by accelerometers and by gyroscopes are preferred. In fact, gyroscopes are more sensitive to the rotational movements of the mandible actuated by its own muscles under the direct control of the brain than accelerometers, which will show the movements of the head that the tracheal traction may produce. In addition to the respiratory movements, in the case of central activation, an isolated signal with a large amplitude will be measured. However, the movement imposed by the tracheal traction is a movement damped by the elasticity of the tissue, which connects the mandible to the rest of the head, and can therefore transmit the movement passively. It is therefore a relatively imperceptible reflex of the spinal drive, that is to say the diaphragm that produces the tracheal traction, while the antagonist/agonist muscles of the mandible, in particular through the action of the driving branch of the trigeminal nerve directly from the brain (i.e., the trigeminal drive), impart direct movement. Gyroscopes are well able to measure the rotational movements of the mandible produced by the muscles of the mandible, and therefore the rotational movements of the mandible are the result of the direct action of the brain on the mandible. Thus, combining the signals from the accelerometer and from the gyroscope can improve the detection of the origin and nature of jaw movement, and therefore improve the determination of whether a person is sleeping.

Preferably, the sensing unit comprises a magnetometer adapted to measure the movement of the subject's head and/or mandible, the device or unit comprises a third input for receiving a third time stream of measurement signals from the magnetometer, the analysis unit being adapted to integrate the measurement signals from the magnetometer with the third signal. The use of a magnetometer makes it possible to determine the absolute position of the head and mandible.

Preferably, the sensing unit comprises an oximeter and/or a thermometer and/or an audio sensor and/or an electromyographic unit and/or a pulsed photoplethysmograph; the identification device or unit comprises a fourth input and/or a fifth input and/or a sixth input and/or a seventh input and/or an eighth input for receiving a fourth time stream and/or a fifth time stream and/or a sixth time stream and/or a seventh time stream and/or an eighth time stream of measurement signals from the oximeter, the thermometer, the audio sensor, the electromyographic unit, the pulsed photoplethysmograph, respectively; the analysis unit is adapted to combine the measurement signals from the oximeter, the thermometer, the audio sensor, the electromyographic unit, the pulsed photoplethysmograph, respectively, with the third signal. Then, the identification device or unit is adapted to associate the measurement signals from the oximeter and/or from the thermometer and/or from the audio sensor and/or from the electromyographic unit and/or the pulsed photoplethysmograph with the third signal. These measurement signals from the oximeter and/or the thermometer and/or the audio sensor and/or the electromyography unit enable more measurement signals to be taken into account and thus the sleep disturbance to be analyzed more reliably.

Preferably, the first predetermined range consisting of frequencies is between 0.15 Hz and 0.60 Hz (including 0.15 Hz and 0.60 Hz), the recognition unit is suitable for identifying the first signal within a time period of at least two breathing cycles of the subject, and the second predetermined range consists of a value that is an amplitude value of the mandibular rotational movement. This value is, for example, an amplitude of the order of 1/10 mm (i.e., based on normal breathing). The frequency range between 0.15 Hz and 0.60 Hz (including 0.15 Hz and 0.60 Hz) characterizes a situation in which the subject's head is quasi-immobile, thus reflecting a situation in which the subject is sleeping or falling asleep.

Preferably, the analysis unit is adapted to identify in the third signal those signals characterizing a rotation of the head around at least one axis extending through the subject's head in the first time stream and the second time stream. During sleep, a rotation of the head will typically be accompanied by arousals, micro-arousals or cortical and/or subcortical activations and is indicative of a sleep disturbance.

The present invention further provides a method for automatically detecting sleep stages from mandibular rotational movement data preferably recorded by means of a gyroscope. The method may be a machine learning based method according to one or more embodiments as described herein. The method preferably comprises the following steps:

- providing sampled rotational motion data from at least one subject; the sampled data may be provided by one or more sampling and processing methods as described herein;

- Feed the provided data to a machine learning classifier to generate a prediction score;

-Determine the sleep stage based on the generated score.

It should be understood that preferred embodiments of other methods described in this specification are also preferred embodiments of the automatic sleep or sleep stage detection method. Data from this method can be used as input to other methods or devices, which can be therapeutic in nature.

In some embodiments, sleep stages may include the following categories (classified by increasing level of complexity):

(1) 2-category (i.e., binary) scoring to detect the subject’s wakefulness or sleep state;

(2) a 3-category score for classifying sleep stages including wakefulness, non-REM sleep stages, or REM sleep stages of the subject;

(3) a 4-category score for classifying the subject's sleep stage into wakefulness, light sleep (N1 and N2) stages, deep sleep (N3) stage, or REM sleep stage;

(4) A 5-category score for classifying all sleep stages including the subject's wakefulness, N1 sleep stage, N2 sleep stage, N3 sleep stage, and REM sleep stage.

An exemplary method for implementing automatic sleep stage detection with 3-category scoring is discussed in Examples 18 and 19.

In addition to detecting sleep-related disorders, the systems and methods described herein can also be used for the following exemplary applications: sleep stage detection and/or sleep quality monitoring of healthy subjects, elderly subjects, or subjects with abnormal sleep patterns. Detection of sleep disorders (whether clinical or psychological) can allow treatment to be customized or meet the needs of the subject. In addition, studying the impact of sleep behavior on clinical outcomes in chronic diseases may allow new insights into the disease and on the efficacy of treatments to be obtained.

In addition, the systems described herein may also be used in combination with other systems or methods. These systems may optionally be therapeutic in nature, such as respiratory devices (CPAP, BiPAP, adaptive support ventilation), mandibular advancement braces, and oral devices, whether percutaneous or implanted, for stimulating nerves and/or muscles, devices for correcting posture and/or position of the body and/or head during sleep. In some embodiments, an alarm may be coupled to the system, or the system may be connected to a device having an alarm function or providing a device having an alarm function.

Additionally or alternatively, the present invention may be described by the following numbered embodiments. In these numbered embodiments, the term "combination" is equivalent to the term "system" unless the context clearly indicates otherwise.

Embodiment 1. A combination comprising a sensing unit and a device for processing data (e.g., a processing unit), the data being related to disturbances that may occur during a subject's sleep, the sensing unit comprising an accelerometer and a gyroscope, the accelerometer being suitable for measuring movements of the subject's head and/or mandible, the gyroscope being suitable for measuring movements of the subject's mandible; the sensing unit being suitable for generating a measurement signal based on the measurements achieved, the device comprising a first input and a second input, the first input and the second input being used to receive a first time stream of measurement signals from the accelerometer and a second time stream of measurement signals from the gyroscope, respectively, characterised in that the device comprises an identification unit, the identification unit being suitable for identifying a first signal and a second signal in the first time stream and the second time stream, the frequency of the first signal being within a first predetermined frequency range, and the value of at least one intrinsic feature characterising the movement of the head and/or mandible being within a first predetermined frequency range in the second signal. within a second predetermined range consisting of a plurality of values; the first predetermined frequency range and the second predetermined range are respectively composed of a value and a value of the frequency of the movement of the subject's head and mandible characterizing the subject's sleep state; the recognition unit is suitable for generating a trigger signal after observing that the first signal and the second signal already identified in the first time stream and the second time stream exist for a first predetermined time period; the recognition unit is also suitable for identifying a third signal in the first time stream and the second time stream after it has generated the trigger signal, in which the frequency and/or value of the at least one inherent feature in the third signal represents the movement of the subject's mandible and/or the change in position of the head; the recognition unit is connected to an analysis unit, which is suitable for being activated under the control of the trigger signal; the analysis unit is also suitable for comparing the third signal with a profile characterizing frequencies and/or values associated with sleep disturbance and generating a result of the comparison.

Embodiment 2. A combination according to embodiment 1, characterized in that the sensing unit comprises a magnetometer adapted to measure the movement of the subject's head and/or mandible, the device or unit comprises a third input for receiving a third time stream of measurement signals from the magnetometer, and the analysis unit is adapted to combine the measurement signals from the magnetometer with the third signal.

Embodiment 3. According to the combination of embodiment 1 or 2, it is characterized in that the sensing unit includes a blood oximeter and/or a thermometer and/or an audio sensor and/or an electromyography unit and/or a pulsed light plethysmograph; the identification device or unit includes a fourth input and/or a fifth input and/or a sixth input and/or a seventh input and/or an eighth input for receiving a fourth time stream and/or a fifth time stream and/or a sixth time stream and/or a seventh time stream and/or an eighth time stream of measurement signals from the blood oximeter, the thermometer, the audio sensor, the electromyography unit, and the pulsed light plethysmograph, respectively; and the analysis unit is suitable for integrating the measurement signals from the blood oximeter, the thermometer, the audio sensor, the electromyography unit, and the pulsed light plethysmograph, respectively, into the third signal.

Embodiment 4. The combination of any one of embodiments 1 to 3, characterized in that the first predetermined range of frequencies is between 0.15 Hz and 0.60 Hz, and the recognition unit is adapted to recognize the first signal within a time period of at least two breathing cycles of the subject.

Embodiment 5. The combination of any one of embodiments 1 to 4, characterized in that the second predetermined range of values includes at least one head movement amplitude value representing a position change of the head.

Embodiment 6. The combination according to any one of embodiments 1 to 5, characterized in that the analysis unit is adapted to identify in the third signal those signals characterizing in the first time stream and/or the second time stream a rotation of the head around at least one axis extending through the subject's head.

Embodiment 7. A combination according to any one of embodiments 1 to 6, characterized in that the recognition unit is suitable for identifying changes in the position of the mandible and the head of the subject in the first signal stream and the second signal stream, and the analysis unit is suitable for removing at least one feature from the motion signal stream for identifying information representing the motion.

Embodiment 8. A combination according to any one of embodiments 1 to 7, characterized in that the processing device is suitable for applying pre-processing to the first time stream and/or the second time stream by applying a band pass filter and/or a low pass filter and/or an exponential activity average and/or a calculation of the entropy of the frequency of the signal to the first time stream and/or the second time stream.

Embodiment 9. A combination according to embodiment 7 or embodiment 8 when dependent on embodiment 7, characterized in that the analysis unit is adapted to verify whether during a second time period, in particular a time period of 30 seconds, the at least one feature of the information used to identify the movement has a value characterizing a sleeping state or a value characterizing a waking state, and if the at least one feature of the information used to identify the movement and removed from the received analysis signals of the first and second time streams has a value describing a sleeping state or a value describing a waking state, the analysis unit is adapted to generate a first data item representing a sleeping state or a waking state.

Embodiment 10. A combination according to any one of embodiments 7, 9 or embodiment 8 when dependent on embodiment 7, characterized in that the analysis unit is suitable for verifying whether the frequency and/or at least one feature of the information used to identify the motion and removed from the analysis signals of the first received stream and the second received stream has a value representing the N1 sleep state or a value representing the REM sleep state during a second time period, in particular a time period of 30 seconds, and if the frequency and/or at least one feature of the information used to identify the motion and removed from the analysis signals of the first received stream and the second received stream has a value representing the N1 sleep state or a value representing the REM sleep state, the analysis unit is suitable for generating a second data item representing the N1 sleep state and a third data item representing the REM sleep state.

Embodiment 11. According to the combination of embodiments 7, 9 or 10, it is characterized in that the analysis unit is suitable for verifying whether the at least one feature used to identify the information characterizing the movement and removed from the analysis signal of the first received stream and the second received stream has a value characterizing the N2 sleep state or a value of the N3 sleep state during the second time period, especially within the time period of 30 seconds, and if the at least one feature used to identify the information characterizing the movement and removed from the analysis signal of the first received stream and the second received stream has a value representing the N2 sleep state or a value representing the N3 sleep state, the analysis unit is suitable for generating a fourth data item representing the N2 sleep state and a fifth data item representing the N3 sleep state.

Embodiment 12. A combination according to any one of embodiments 1 to 11, characterized in that the analysis unit is suitable for verifying whether at least one inherent feature of the analysis signal of the first received stream and the second received stream has a level representing cortical activity or subcortical activity during a third time period, in particular, a time period between 3 seconds and 15 seconds, and if the at least one inherent feature of the analysis signal of the first received stream and the second received stream has a level representing cortical activity or subcortical activity, the analysis unit is suitable for generating a sixth data item representing cortical activity or subcortical activity.

Embodiment 13. A combination according to any one of embodiments 1 to 12, characterized in that the analysis unit is suitable for verifying whether at least one inherent feature of the analysis signal has a level characterizing obstructive apnea, obstructive hypopnea, respiratory effort associated with arousal, central apnea, central hypopnea, and if the at least one inherent feature of the analysis signal of the first time stream and the second time stream has a level describing obstructive apnea, obstructive hypopnea, respiratory effort associated with arousal, central apnea, or central hypopnea, the analysis unit is further suitable for generating a seventh data item, an eighth data item, and a ninth data item, and the seventh data item, the eighth data item, and the ninth data item represent obstructive apnea, hypopnea, respiratory effort associated with arousal, central apnea, or central hypopnea.

Embodiment 14. A combination according to any one of embodiments 1 to 13, characterized in that the recognition unit is adapted to recognize frequency values and/or values of at least one intrinsic feature in the first time stream and the second time stream, the at least one intrinsic feature showing variability not observed during the sleep state, and to generate a neutralization signal when such variability is observed and to provide the neutralization signal to the analysis unit to neutralize the variability.

Embodiment 15. A combination according to any one of embodiments 1 to 14, characterized in that the analysis unit is adapted to verify whether at least one intrinsic feature of the analysis signals of the first time stream and the second time stream has increased by more than at least 1 mm, when the movement is phasic, at a frequency established in the range of 0.5 Hz to 5 Hz during at least three breathing cycles, or by more than 1 mm in a continuous, tense manner for at least 2 seconds, and during such verification a tenth data item representing bruxism is generated.

Embodiment 16. A combination according to any one of embodiments 1 to 15, characterized in that the analysis unit is adapted to capture one or more values in the first time stream and in the second time stream, the one or more values allowing access to the calculation of the loop gain, the calculation of the muscle gain for mobilizing the mandible during apnea or hypopnea or a period of effort, the passive collapsible point from after activation and/or the arousable point from before activation.

Example

Example 1

In a first example, see Figure 1. Figure 1 shows a system according to the invention. The system comprises a sensing unit 1 and a device 10 for processing data, preferably a processing unit, the data being related to disturbances that may occur during the sleep of a subject. The sensing unit comprises an accelerometer 2, which is suitable for measuring movements of the subject's head and/or mandible, preferably in three dimensions. The sensing unit further comprises a gyroscope 3, which is suitable for measuring rotational movements of the subject's mandible, preferably in three dimensions. According to a preferred embodiment, the sensing unit 1 further comprises a magnetometer 4 (in particular in the form of a compass) and/or an oximeter 5 and/or a thermometer 6 and/or an audio sensor 7 and/or an electromyographic unit 8 and/or a pulsed photoplethysmograph 9. Other sensors, such as a perspiration sensor or a nasal pressure sensor, may also form part of the sensing unit. The pulsed photoplethysmograph functions by transmission or by reflection and allows access to the calculation of the frequency of the pulses and the calculation of the changes in arterial tension.

The sensing unit preferably has small dimensions, for example at most 5 cm long, 2 cm thick and 1 cm high, so as not to interfere with the normal sleep of the subject. The sensing unit preferably has very small overall dimensions, is lightweight and flexible, enabling good ergonomics. The signals generated by the sensing unit are very suitable for decoding using artificial intelligence. The diagnostic power of the measurements obtained by the sensing unit is comparable to that of a complete polysomnographic recording. Movement of the mandible may occur preferentially on an axis, for example on the anterior-posterior axis, while the subject's head is turned to the right. Movements on other axes may similarly be measured. For reasons of hygiene, the sensing unit is preferably intended to be used only once, but it can of course be repaired and reused.

The position of the head is preferably determined based on the values measured by the accelerometer 2 along three axes. When the accelerometer measures the value of the acceleration relative to the gravity of the ground, these measurements are preferably combined over time in order to obtain the position of the head, if there is no initialization phase during the application of the sensing unit to the head of the person, then the position of the head will be a relative position. For example, the position can be expressed according to the values of the Euler angles of pitch, roll and yaw, or again by divisions of 15°. The position of the head can also be expressed in the following terms: standing, lying down, left, right, behind.

The following table shows various angle values and the resulting head positions:

Pitch Roll yaw Location 80° 0° 10° upright 10° 10° 70° Lie down with your head facing left 20° 0° 15° Lie down with your head facing backwards

A magnetometer 4 will be added to sense the orientation of the head, in particular when movement occurs perpendicular to the gravity vector. Combining the values measured by the accelerometer and magnetometer enables calculation of the distance of movement and therefore obtaining an absolute value of the position of the head.

Regarding the movement of the head, the movement of the mandible is preferably measured in three axes by means of measurements from an accelerometer 2. The movement of the mandible is also measured by means of a gyroscope 3.

The movements of the head and mandible and the resulting position changes are of different kinds. For the mandible, the movements are, for example, rotational movements at the respiratory rate. However, in the case of bruxism or chewing or in the case of oral motor disorders, lateral movements may occur during sleep, and there the mandibular condyle undergoes a rotation in the glenoid cavity of the temporomandibular joint, but these are not the same axes as in the case of respiratory movements.

For the head, the results of the movement are random, i.e. after activation it is not possible to predict the position that the head will occupy at the end of the movement. The amplitude of the movement and the amplitude of the change in position have different values. Therefore, if the amplitude of the movement of the head is high, the change in the position of the mandible measured by the gyroscope will not be studied, because if this is the case, the subject will be awakened and no information about the subject's sleep disturbance will be obtained. Small amplitudes of the movement of the mandible captured by the gyroscope are observed when they originate from respiratory movements. The change in the yaw angle is associated with the head and indicates a rotation of the head from left to right. The change in the pitch angle is associated with the flexion or extension of the head, in addition to the other parameters used to provide information about the movement of the mandible. These values of the captured signals will be analyzed with the help of an analysis unit, as described below.

Respiratory as well as non-respiratory movements can cause jaw movement. Thus, movement of the head while a person is sleeping may cause jaw movement. Jaw movement can be caused by tracheal traction or by the person's brain. Tracheal traction is the traction exerted by the chest on the person's head. This traction is at the person's breathing frequency. Therefore, if the head moves at the breathing frequency, the mandible, which is attached to the head, will follow the movement exerted by the head at the breathing frequency. This is a passive movement that follows the movement of the head. In the absence of head movement, jaw movement can likewise be directly and actively controlled by the brain. When the brain controls jaw movement, the muscles of the mandible are directly stimulated. Therefore, it is useful to be able to clearly distinguish between jaw movement controlled by the brain and jaw movement controlled by tracheal traction.

A distinction is made between isolated mandibular movements (IMM) that occur when the brain is activated (e.g., at the end of a period of respiratory effort, during coughing or spitting, or when speaking again in sleep) and respiratory mandibular movements (RMM) that are caused by the subject's breathing. There are also mandibular movements caused by grinding or chewing. RMM-type mandibular movements are directly controlled by the subject's brain and do not cause movement of the head. RMM-type movements can also be produced by tracheal traction, which will then be combined with the movement of the head at the respiratory rate. When RMM-type movements stop, normalize, or start, it is useful to observe whether the head moves on this occasion with the help of measurements achieved by an accelerometer. Grinding-type movements usually occur after an activation that causes head movement and is indicated by the accelerometer, because it does capture large-amplitude movements that contrast with the relatively fine rotational movements of the mandible that are clearly shown by the gyroscope.

The device 10 for processing data related to sleep disturbance according to the present invention comprises a first input 11-1 for receiving a first time stream F1 of a measurement signal from an accelerometer 2, i.e., measured acceleration data. The device 10 comprises a second input 11-2 for receiving a second time stream F2 of a measurement signal from a gyroscope 3, i.e., measured rotational motion data. The device 10 may also comprise a third input 11-3 for receiving a third time stream F3 of a measurement signal from a magnetometer 4, i.e., magnetic field data. When the sensing unit further comprises an oximeter, the identification device will comprise a fourth input adapted to receive a fourth time stream F4 of a measurement signal from the oximeter, i.e., oximeter data. When the sensing unit further comprises a thermometer, the identification device will comprise a fifth input adapted to receive a fifth time stream F5 of a measurement signal from the thermometer, i.e., thermometer data. When the sensing unit further comprises an audio sensor, the identification device will comprise a sixth input adapted to receive a sixth time stream F6 of a measurement signal from the audio sensor, i.e., audio data. When the sensing unit also comprises an electromyographic unit, the identification device will also comprise a seventh input adapted to receive a seventh time stream F7 of measurement signals from the electromyographic unit, i.e. electromyographic data. When the sensing unit also comprises a pulsed photoplethysmograph, the identification device will also comprise an eighth input adapted to receive an eighth time stream F8 of measurement signals from the pulsed photoplethysmograph, i.e. photoplethysmography data. In other words, the measurement data from the various sensors are sent from the sensors to the analysis unit via the data link.

The various inputs need not be physically distinct, as the various streams may be time division multiplexed and/or each stream carried by a carrier wave of a different frequency. Thus, the various input streams may be sent over a single data link.

The device comprises a data analysis unit, the data analysis unit comprising an identification unit 12, the identification unit 12 is adapted to identify a first signal and a second signal in a first time stream F1 and a second time stream F2, the frequency of the first signal being within a first predetermined range consisting of frequencies, the value of the second signal being within a second predetermined range consisting of values, the first predetermined range consisting of frequencies and the second predetermined range consisting of values being composed of frequencies and values of movements of the subject's head and mandible, respectively, characterizing the subject's sleep state. When the sensing unit comprises a magnetometer 4, the identification unit 12 will also be adapted to identify a third signal in a third time stream F3, the value of the third signal being within a third predetermined range of values of the orientation of the subject's head, such as can be observed during sleep. The identification unit is adapted to generate a trigger signal after observing that the first signal and the second signal, which have been identified in the first time stream and the second time stream, exist for a first predetermined period of time. The identification unit is also adapted to identify a third signal in the first time stream and the second time stream after it has generated the trigger signal, the frequency and/or value of the third signal characterizing the movement of the subject's mandible and/or the change in position of the head. The identification unit is connected to an analysis unit 13, which is adapted to be activated under the control of the trigger signal. The analysis unit is further adapted to compare the third signal with a profile characterizing frequencies and/or values associated with a sleep disturbance and to generate a result of the comparison.

Specifically, the identification unit may be included in a data analysis unit, which also includes a memory unit. The memory unit is configured to store N mandibular motion categories, wherein N is an integer greater than 1, wherein at least one of the N mandibular motion categories represents a sleep disordered breathing event. Each j-th (1≤j≤N) mandibular motion category includes a j-th rotation value set, and each j-th rotation value set represents at least one rate, rate change, frequency and/or amplitude of mandibular rotation associated with the j-th category. In addition, each j-th mandibular motion category optionally includes a j-th acceleration value set and/or a j-th magnetic field data value set. The data analysis unit includes a sampling element, which is configured to sample the measured rotational motion data and optionally the measured acceleration data and/or the measured magnetic field data during a sampling period, thereby obtaining sampled rotational motion data and optionally sampled acceleration data and/or sampled magnetic field data. The data analysis unit is configured to derive a plurality of measured rotation values from the sampled rotational motion data; and optionally derive a plurality of measured acceleration values and/or measured magnetic field values from the sampled acceleration data and/or the sampled magnetic field data. The data analysis unit is also configured to match the measured rotation values with N mandibular motion categories. Optionally, the data analysis unit is also configured to match the measured acceleration values and/or magnetic field values with N mandibular motion categories. Therefore, sleep disordered breathing events are effectively detected.

Regarding the data link: The device and the sensing unit preferably communicate with each other wirelessly, but it goes without saying that a cable connection is equally possible. The device is preferably part of a computer located in a data processing center. Wireless communication is achieved, for example, by means of a telephone communication network, and the sensing unit is, for example, equipped with a Bluetooth system enabling it to communicate with the telephone. Thus, the stream of measurement signals generated by the sensing unit will be transmitted to the device.

The invention is based on the fact that it has been observed that the movements of the mandible depend not only on the movements of the chest, as indicated in the literature, but also on direct control from the neural centers of the brain, which control the muscles attached to the mandible, whose function is to position the mandible. It has been observed that the position of the head, and especially the changes in its position during sleep, can stop all mandibular movements or start them in a completely independent way from the movements of the chest. That is, the mandibular movement can follow in the presence of the movement of the chest only when the position of the head allows it and does not fix it. Therefore, the movement of the head can act on the movement of the mandible or paralyze it, in the sense that it is nothing more than an epiphenomenon of the brain activation that marks micro-arousal or arousal, and can have other effects on the movement of the mandible.

The movements of the head actually affect the permeability of the upper airway by applying a squeezing force when the upper airway is more contractile during sleep, or by activating/deactivating the muscle motor units of the upper airway. These movements of the head during sleep change the permeability of the upper airway and must be understood and superimposed on the movements of the mandible in a timely manner. Therefore, these movements of the mandible can be correctly analyzed and then interpreted based on the changes in breathing control starting from the air flow generated by the sleeping subject. In other words, sensing and analyzing the jaw movement will take into account the brain control, taking into account the position of the head and the changes in the position of the head during sleep, whether in the case of micro-arousals or arousals, in order to position or reposition the mandible by activating/deactivating the muscles attached to the brain. In addition to the brain activation, the movement of the position of the head at the respiratory frequency will be generated by the traction of the trachea, while the jaw movement at the same frequency is directly determined by the neural center.

By actuating the mobile bone formed by the mandible in the manner of a lever, brain control seeks to strengthen the upper airway by activating the tongue muscles and the muscles of the pharynx attached to the tongue to prevent apnea. To do this, brain control relies on muscles that increase or decrease at the respiratory rate during sleep, opening or closing the mouth. Brain control can also act on the muscles that push the mandible forward (also at the respiratory rate), or even in a combined manner, these combined muscle groups involve movement in different directions.

Changes in the position of the head during sleep are often accompanied by arousals or micro-arousals, which can also be recorded, for example, by electrodes placed on the scalp and record the activity of the cortex of the brain. When there is anyway a movement of the head in the case of jaw behavior modification, the scalp electrodes sometimes do not record activation. The reason for this is that the activation is already held subcortically in the brain stem and is sometimes completely autonomous. These movements of the head are performed completely independently of the chest movements.

Because the position of the head is no longer aligned with that of the body, or because the change in the position of the head is an epiphenomenon of a spontaneous or involuntary flip under the control of the neural centers, the analysis of the jaw movements in the vertical plane and in the horizontal plane, depending on the position of the head, can provide information about the level of respiratory effort, in particular its amplitude, by creating a twist of the neck, with that control performed by the neural centers of the brain in the case of a change in the resistance to the flow of air through the upper airways. When the control from the neural centers increases, the respiratory event is considered an increase in effort, and when the control from the neural centers decreases, the respiratory event is considered central. The brain control that enables the organism to leave apnea must activate the jaw lever upward in the vertical plane and forward in the horizontal plane, ideally with the head aligned axially with the body in order to prevent any compression of the upper airways. This (micro)arousal itself is identified by an isolated large jaw movement (IMM), and the duration of the (micro)arousal, whether respiratory or non-respiratory, is measured and clearly distinguished from the subsequent jaw movement.

Example 2

In further examples, see Figures 2A and 2B.

In a state controlled by the brain, during sleep, the results of the analysis of the measured data stream from the sensing unit provide information, and the change in the position of the head represented by the signal from the accelerometer is usually a sign of its state change. Figure 2A + Figure 2B show the stream during the change of position of the head of a person lying in bed. This movement can never be superimposed on the movement of the mandible in wakefulness and therefore also not on the movement of the mandible in the conscious state during chewing, phonation or swallowing as studied by practitioners in this field except for the movements reserved for sleep medicine. The latter involves chewing, phonation and swallowing problems studied in dentistry, stomatology, maxillofacial surgery, orthodontics, orthodontics, speech correction, etc. in conscious subjects who are not in a sleeping state.

Fig. 2A shows the change of the head from the first position to the second position from left to right, in the first position the head turns to the left, in the second position the head turns to the right. Then the change to the third position is seen, in which the head turns to the left again. The first time stream F1 generated by the accelerometer is related to the three axes (Fx, Fy, Fz) of the three dimensions that realize the measurement. The second time stream F2 generated by the gyroscope also involves these three axes. At the moment of the head turning, it can be clearly seen that the two streams have a peak with a high amplitude. It can also be seen that when the head is in the first position, the first time stream F1 and the second time stream F2, especially in the vertical direction y of the first time stream F1, have a larger variable amplitude, which represents an increased brain control state, represented by reference mark 1, and is variable in terms of control intensity. In addition, this is also seen in the stream Ft showing the movement of the chest. Therefore, the analysis unit can infer from the stream that the person is showing an increased and variable breathing effort.

When the rotation of the head has occurred and the head is in the second position, it can be seen that the amplitude of the first time stream F1 has decreased significantly, as has the amplitude of the second time stream F2. The level of stream F1 has decreased, which indicates that the mouth has opened, as shown in reference mark 2. It can also be seen that the fifth time stream F5 has decreased, which may lead to a loss of oxygen flow (reference mark 3). It can also be seen in the second time stream F2 that the amplitude has decreased, which indicates a loss of brain control of the amplitude, as shown in reference mark 4. All of this indicates that the amplitude of the effort has decreased and breathing has been affected (see the fifth time stream F5), which also causes the brain to activate and generate commands that cause a new change in the position of the head, with the head turning to the left. After this, it can be seen in the second time stream F2 that the amplitude has become larger and the fifth time stream F5 has increased. Therefore, it will be found that brain control tends to normalize breathing.

Fig. 2B shows that even small changes in the position of the head are caused by brain control. Fig. 2B shows a change in which the head has rotated slightly to the right. As indicated by arrow 1, the first time stream F1 first shows that the brain control state has increased and a breathing effort has been generated. It can be seen that when the head changes position, the accelerometer (F1) shows an increase in amplitude and frequency representing brain activation, indicated by reference mark 2. In the eighth time stream F8 (EEG) and the seventh time stream F7 (EMG), a time period of 30 seconds of brain activation can be clearly seen, zoomed in here (reference mark 2). It can then be seen that the level of the first time stream F1 (reference mark 3) shows a brain control state with reduced amplitude and the mandible has been raised (mouth has closed).

The technology employed by the system according to the invention unexpectedly and unpredictably provides information about the nature of the jaw movement during sleep, its central origin, the control of the neural centers that must strengthen the pharynx to maintain ventilation and thus oxygen supply to the subject, while during sleep the head end must ideally remain aligned with the body, in particular with the torso. Therefore, the jaw movement must be interpreted in terms of the position of the head and its changes, otherwise it will not be understood why the jaw movement stops or starts or changes amplitude during sleep.

Example 3

In a third example, see Figures 3A and 3B.

The technology provided herein is applicable to the detection of bruxism. Known diagnosis of bruxism utilizes the electromyogram of the masseter and anterior temporalis muscles and possible anterior temporalis muscles during polysomnography in the laboratory, and the examination must also include audio-video recording. Since the demand for sleep recording is disproportionate to the recording capacity of the sleep laboratory, such examination is expensive, laborious and somewhat difficult to achieve. The recording is achieved in one night, and its laborious nature usually prevents it from being repeated. In addition, in order to track bruxism, it must be possible to record at multiple nights, because bruxism may not be systematically reproduced every night and remain intermittent. Therefore, it is necessary to implement it in the home of the relevant subject, under real-life conditions and without interfering with the natural progression of sleep. Results must be given quickly to optimize the control of bruxism and verify the therapeutic effect.

Currently, bruxism is not detected at home because there is no technical solution to achieve this. The proposed solutions such as surface electromyography of the masseter or anterior temporalis muscles do not ensure the diagnosis of the pain. In fact, the recording of the electromyographic activity of the masseter or anterior temporalis muscles alone is affected by parasitic movements during the night or because the fatty medium on the muscles prevents the capture of their electromyographic (EMG) activity. Video recording allows the laboratory to verify the movement of the mandible and the resulting electromyographic activity corresponding to bruxism.

The technical solution proposed by the present invention includes recording the mandibular movement with the help of a sensing unit, preferably recording the mandibular movement on the three main axes of the movement of the mandible in space, and then performing an algorithmic analysis of the signal with the help of an analysis unit. This analysis makes it possible to identify mandibular movements that develop specifically and exclusively during the onset of bruxism and are established by detecting RMMA (rhythmic muscle masseter activity), that is, there is a phased but sometimes only tonic activity during the surface electromyography of the masseter muscle. The signal stream generated by the sensing unit is analyzed on three axes, which also makes it possible to capture lateral movements that may be imposed during the grinding of the teeth and contribute to the wear of the enamel. The mandibular movement (called bruxism) is the result of the joint action of agonist and antagonist muscles, which involve not only the lifting group of the mandible (such as the anterior temporalis muscle), but also the medial and lateral hypoglossal muscles and pterygoid muscles.

Fig. 3A + Fig. 3B show the streams captured by the capture unit during a bruxism visit. The EMG activity of the recorded muscles (see streams F7D and F7G) has been validated as contributing to jaw movements. The typical features of the electromyographic activity of the masseter and/or anterior temporalis muscles are reflected in the jaw movements, which are also diagnostic for bruxism. The latter are superimposed in the form of modulated signals on the (sustained) tense or (rhythmic) phasic bruxism myoelectric bursts that generated them. The duration of a cycle or a burst can be calculated.

The effort period preceding the onset of bruxism (indicated by arrow 1) can be easily identified by analysis of the jaw movements and by a brief awakening (indicated by arrow 2) accompanied by cortical or only voluntary subcortical activation. (Whether cortical, e.g. reflected exclusively in changes in cortical wave frequency on the EEG, as indicated by the eighth time stream F8, or subcortical and not visible on the EEG) The activation is well marked by the preceding jaw movement and is described in the literature as generally preceding the onset of bruxism. It should be noted that the masseter phase and/or tonic activity peak are simultaneous with the extreme positions of the jaw movement, clearly verifying the relationship between muscle recruitment and movement of the active bones of the jaw. In FIG3A , it can be seen that there is a period of effort in the first time stream F1, indicated by arrow 1, followed by activation, indicated by arrow 2, and in turn followed by the movement of the mandible caused by bruxism, indicated by arrow 3. FIG3B is a magnified view of the 10 second period indicated by arrow K in FIG3A to the upper right. FIG. 3B shows the synchronization between the activity in the EMG of the right masseter (F7D) and left masseter (F7G) muscles and the molar mandibular movement.

It can be seen here that the recovery activity of the seventh time stream F7 (EMG) of the right masseter (F7D) is synchronized with the recovery activity of the left masseter (F7G) and the recovery activity of the mandibular movement caused by bruxism. The figure clearly shows that after a period of effort, the respiratory frequency of the position of the mandible with abnormal amplitude has changed. After the movement marking cortical activation, four high-frequency (1Hz) rotational movements corresponding to the onset of bruxism, there are accompanied by head movements in F1Z and large movements on the gyroscope F2X. After this, a period of effort appears again.

The movements of the head and the mandible analyzed by their intrinsic characteristics (that is, in particular the frequency characteristics and morphological characteristics of the signal flow) can be distinguished according to their generation mechanism and sorted in chronological order. These characteristics can be observed by analyzing, for example, the amplitude, area or slope of the measurement signal. These characteristics are, for example:

Movement associated with respiratory effort, then

Movements associated with transient cortical or subcortical activation, followed by

• Movements associated with a grinding or chewing movement that can be clearly distinguished, such as, for example, the number of bursts during a grinding cycle, the length of the cycle between two bursts, or the duration of the bursts.

Mandibular movements are produced by the agonist/antagonist action of the muscles for raising and lowering the mandible. The latter are directly controlled by the nucleus of the cerebral nerve center of the trigeminal nerve drive branch. Mandibular movements can be sensed here by the change in the angle assumed by the mandible during its movement relative to a plane, for example during its vertical movement relative to the horizontal plane.

Even if the chest movement continues, the mandibular movement may start or stop only if the position of the head changes. Changes in the head position always occur simultaneously with cortical or subcortical micro-arousals and therefore interfere with the control of the brain's neural centers. Even if there are no longer abdominal chest movements, that is, even if the diaphragm muscles that act on the expansion of the chest and abdomen during inspiration controlled by the spinal nerves no longer work or have stopped, the mandibular movement may continue at the respiratory rate during sleep. The mandibular movement can then be imposed in the form of a front-to-back or back-to-front movement in another plane (for example the horizontal plane), that is, in a plane other than the plane of craniocaudal traction, thereby affecting the tracheal traction.

In the first time stream provided by the accelerometer, and likewise in the second time stream provided by the gyroscope, a tonic phase movement of the jaw position at the respiratory frequency upwards, i.e. in the direction opposite to that which would be exerted by the traction generated by the tracheal traction, can be seen. This upward and forward movement is produced by the contraction of the anterior temporalis and masseter muscles and pterygoid muscles, respectively, in particular by the upper muscle group.

When respiratory effort begins and when the amplitude of the jaw movement will increase due to the increase in central respiratory control, the direction imposed on the jaw movement may also be in a plane other than the vertical plane, which is the plane of traction. This is due to the effect of certain muscle groups that are recruited more than other muscle groups, such as the pterygoid muscles that are recruited more than the hypoglossal muscles. Movement at the respiratory rate may occur in a more horizontal direction that will be captured by the inertial unit. The inertial unit includes an accelerometer and a gyroscope. In fact, if the effort is monitored only in the vertical plane, the effort cycle may escape the signal analysis. Movement may also occur mainly in one direction (vertical or horizontal) and not in the other.

The shape of the respiratory motion, and in particular its acceleration slope, varies depending on the muscle groups recruited. During vertical motion, when the masseter muscles are active, the direction of motion during inspiration is upward (opposite to that observed when antagonist lower muscle activity predominates) and results in a reduction in motion, which generates a change in the motion waveform.

This analysis of the stream provided by the sensing unit makes it possible to verify the fact that the movement of the mandible during inspiration is downward when the activity of the lowering muscles is dominant, and upward when the activity of the lifting muscles is dominant. This information is obtained via the analysis of the variations in the captured velocities and accelerations. This makes it possible to assess the level and nature of the response developed by the subject to stop the unfolding respiratory event, as well as to assess the greater or lesser recruitment of the muscles that raise the mandible with the task of stabilizing the upper airway.

Example 4

In a fourth example, see FIG. 4 .

The observed flow refers to the identification of four features that describe the behavior of the mandible during the event. These features will enable the understanding of how the subject constructs a respiratory event in a specific sleep stage and a specific position of the head and how the brain will respond to trying to free itself. In addition to the description of the progression of the event, information about its risk of recidivism in both the short and long term can also be identified. For example, when the value of the amplitude of the response relative to the disturbance (called loop gain) is high, that is, when the response to the disturbance is high, these features have a predictive value. Figure 4 shows the loop gain. In this figure, arrow 1 marks the collapsible point on the first time flow F1, that is, the solution of brain control is no longer carried out, so that the mandible is passively lowered under the action of local anatomical constraints (for example, its weight determined by the subject's obesity). Arrow 2 shows the movement of the mandible, and the movement of the mandible is also seen in flow 2 at the same time. At the beginning of arrow 2, the peak-to-peak amplitude of the movement of the mandible is low. Then, thereafter, when the mouth is about to open, the mandible is about to lower, which can be seen at the level of the lowered flow 1, and the peak-to-peak amplitude is about to increase. The level of stream 1 will then reach a level indicated as 3, corresponding to the arousability point, which in turn will be followed by a much larger amplitude peak indicated by arrow 4. This large amplitude movement enables the measurement of the loop gain, which accompanies the closed mouth as shown by the peaks in the first time stream F1 and the second time stream F2, and the highest value that stream 1 will then reach, although the mouth has closed again in the meantime. The loop gain represents the response to the disturbance. It is calculated as the ratio of the difference between the significant points indicated by arrows 4 and 3 and the numerator, and the difference between arrows 3 and 1 and the denominator.

There is a high risk of seeing a repetition of the event in a self-sustaining manner with short apneas, in particular in a central form. Assessment of the muscles of the upper airway, in particular the phasic gain, makes it possible to predict the duration of the event. A low muscle gain indicates a risk of the event continuing for a longer time than when the gain is high. The arousal point, which is the lowest point of the mandibular position just before the activation that terminates the event, also makes it possible to predict the duration of the event. If the position does not drop too much, there is a risk of the event repeating, sometimes cyclically. The influence of anatomical limitations, such as those related to body weight and local accumulation of fatty tissue in the upper airway, can be determined in particular by calculating the position of the mandible (collapse point) based on the values measured by the accelerometer, but also by micro-arousals when the center is still shocked by the latter or by jaw drooping immediately after an arousal.

Example 5

In a fifth example, see FIGS. 5 and 6 .

The stream of measurement signals generated by the sensing unit may include noise that affects the measurement signals, and preprocessing the stream when received by the device may prove useful. The principle of this preprocessing is simply to produce an enhanced signal. Analysis performed by a person skilled in the art has made it possible to know that during the time period with enhanced brain control state, the position of the mandible and therefore the velocity of the mandible and the acceleration of the mandible periodically change by approximately the same value at a frequency of the same order of magnitude as the respiratory frequency, that is to say between 0.15 Hz and 0.60 Hz. It is possible to isolate the signals related to micro-arousals by retaining only the lower frequencies of the respiratory band (for example, by low-pass filtering the measurement signals from the accelerometer and the gyroscope). Figure 5 shows that, by applying this preprocessing, the micro-arousals representing activation are bursty relative to the time period with enhanced brain control state. In the case of each micro-arousal, a clear peak can be seen in the signal. For example, by applying a sixth-order Butterworth filter known in the field of digital signal processing, it is possible to apply this preprocessing.

Conversely, a set-aside period with an enhanced brain control state can be set by filtering one of the captured signals by means of a bandpass filter corresponding to the respiratory frequency band. The result of applying such a filter to the signal from the gyroscope is shown in Figure 6. It can be seen that during effort, the value of the signal is higher.

Features used to identify information in the measurement signal stream are, for example:

- Position of the head and mandible (e.g., roll, pitch, yaw);

- acceleration of the mandible and head along each axis;

-Rotation speed of the mandible and head along each axis;

- the norm of the rotational velocity of the mandible and head about one or more axes (in space, if a vector u has coordinates (x,y,z), its norm is written as: (x ² +y ² +z ² ) ^0.5 );

- norm of the acceleration of the mandible and head along one or more axes;

- the median of the values measured within 10 or 30 seconds or defined by two activations;

- the average of the values measured within 10 or 30 seconds or defined by two activations;

- the maximum of the values measured within 10 or 30 seconds or defined by two activations;

- the minimum of the values measured within 10 or 30 seconds or defined by two activations;

- standard deviation of the values measured within 10 or 30 seconds or defined by two activations;

- exponential moving average of the measured values (with half-lives of 5, 60, 120 and 180 seconds);

- Fourier transformation and integration of the measured values across all frequencies, across the respiratory frequency band (0.15 Hz - 0.60 Hz), across the low frequency band (0 Hz - 0.10 Hz);

- Fourier transformation and identification of the energy maximum frequency or the second energy maximum frequency of the measured values;

- Shannon entropy over a 90 second window of measured values;

- Time offset of the rotational velocity and acceleration signals of the mandible and head and other features in order to take into account the past and the future.

It is likewise possible to combine the above methods with one another.

When features have been identified in the stream of measurement signals, the analysis unit can proceed to analyze them. For this purpose, it will call a random forest type algorithm, for example, using artificial intelligence. Features extracted in this way from the entire series of signal segments for which the polysomnography results are known are injected into the algorithm in parallel with the expected results in order to produce a model that will enable pattern recognition type classification of new segments.

A signal pattern is a specific state of a signal sequence, which can be physically or mathematically visible via parameters. Pattern recognition is the process for identifying (classifying) specific patterns in a signal based on already acquired information or statistical parameters extracted from the signal with the aid of an automatic learning algorithm.

Deep learning is an automated machine learning technique that involves models inspired by the structure of the human brain, known as artificial neural networks. These networks consist of multiple layers of neurons that are able to extract information from data and produce results. The technique works very well for unstructured types of data, such as images, sequences, or biological signals.

Automatic learning (or statistical learning) is a field of artificial intelligence that aims to apply statistical modeling methods that give machines (computers) the ability to learn information from data in order to improve their performance in solving tasks without being explicitly programmed for each of them.

Artificial intelligence (AI) is a set of technologies designed to enable machines to simulate intelligent activities.

For example, the development of these models can be done as follows:

1) 200 subjects were equipped with a sensing unit while they underwent a reference clinical examination in the field of sleep: polysomnography.

2) Then, the signals captured from 40 of these subjects were used to train each random forest model. The signals from the sensing units and a subset of the features obtained after the preprocessing step were injected along with the reference results of the sleep check in the random forest algorithm, and a classification model was generated based on this input data.

3) The remaining subjects are then used in order to validate the model: the signals from the sensing units corresponding to these subjects are injected into the model generated in the previous step, thereby generating results, and those results are compared with those obtained by polysomnography. When the agreement between the results obtained by means of the model and by polysomnography is considered sufficient, the model is considered valid. Otherwise, the development starts again from step 2 of this section.

In order to be able to reliably identify disturbances occurring during the subject's sleep, it is preferably possible to observe that the subject has actually entered a sleep stage. When it has been detected that the subject is actually in a sleep stage, it will then also be possible to establish the subject's sleep stage in order to be able to correctly interpret the signals present in the measurement signal stream. When entering sleep, the mandible will assume a breathing frequency of, for example, between 0.15 Hz and 0.60 Hz. This breathing frequency must be present for a period of several tens of seconds in order to be able to confirm a stable sleep state.

Example 6

In a sixth example, various sleep stages of a subject are discussed. Specifically, Table 1 (included after the following examples) shows various sleep stages of a subject and their relationship to the movement of the mandible and the position and movement of the subject's head. The wake state is essentially characterized by the unpredictable movement of the mandible in this state, while in the subject, in the absence of sleep disturbances, the sleep state is characterized by rotational movement of the mandible at the respiratory frequency. In order to detect the wake state and the sleep state, respectively, the analysis unit will preferably run with an analysis window of 30 seconds and pre-processing the first time stream and the second time stream using a bandpass filter and/or an exponential moving average. In order to extract a profile characterizing the wake state or the sleep state, the level of the standardized mean, for example, will be considered. This level is actually higher in the wake state than in the sleep state.

Also distinguished in sleep are the N1, N2, N3 and REM (Rapid Eye Movement) stages. During the N1 sleep stage, changes in the movement of the mandible at the respiratory rate can be seen in adults, with peak-to-peak amplitude changes lasting a period of time that is usually limited to a few minutes. The position of the head usually remains stable, but the position of the mandible remains unpredictable or may change periodically. In order to detect the N1 sleep stage using a processing device, an analysis window of 30 seconds is preferably used in order to ensure the continuity of the movement. Pre-processing of the first time stream and the second time stream can be used by calculating the entropy of the frequency of the signal. The level of the standardized mean will be considered in the first method as a profile that characterizes this N1 sleep stage, but other methods can be used to improve the analysis accuracy. In the N1 stage, the level of the standardized mean will be higher than in the N2 or N3 stages.

The analysis unit is adapted to verify whether said normalized mean and variance of the amplitude and frequency of the received first and second time streams have a level characteristic of the N1 sleep state during a second time period, in particular a time period of 30 seconds. If said normalized mean and variance of the amplitude and frequency of the received first and second time streams have a level characteristic of the sleep state N1, the analysis unit is adapted to generate a second data item representative of the N1 sleep state.

During the N2 and N3 sleep stages, the brain controls the change in amplitude and/or frequency from N2 to N3 to become lower and lower. Therefore, in normal subjects, there is almost no movement of the mandible or head during these stages. In order to detect the N2 or N3 sleep stage with the help of a processing device, an analysis window of 30 seconds will preferably be used to ensure the continuity of the movement. It is also preferred to use pre-processing with the help of a low-pass filter or a band-pass filter. In the first method, the level of the standardized mean will be considered as an overview for characterizing the N2 sleep stage. In the N2 or N3 stage, the level of the standardized mean will be lower and lower. The level of the standardized median can also be used to identify the N2 or N3 stage or other statistical measurement techniques.

The analysis unit is adapted to verify whether the standardized mean and/or the standardized median of the first time stream and the second time stream received during the second time period, in particular during the time period of 30 seconds, have a level that characterizes the N2 sleep state or the N3 sleep state, respectively. If, for example, the standardized mean and/or the standardized median of the first time stream and the second time stream received have a level that characterizes the N2 sleep state or the N3 sleep state, respectively, the analysis unit is adapted to generate a fourth data item representing the N2 sleep state or a fifth data item representing the N3 sleep state, respectively.

In humans, the REM stage is characterized by unpredictable movements of the mandible. In order to detect this stage with the aid of a processing device, an analysis window of 30 seconds will preferably be used in order to ensure the continuity of the movement. In adults, this type of movement of unpredictable frequency and/or amplitude usually lasts longer in REM than in the N1 stage. The period of this movement during the N1 stage is usually limited to a few minutes. The direction of movement of the mandible position during brain activation is usually negative because the mouth is open. In the REM stage, changes in the movement of the mandible at the respiratory frequency can be seen, where the changes in the peak-to-peak amplitude are not periodic. During the REM stage, the position of the head usually remains unchanged. Detection is performed in a similar way to during N1, with the aim of observing respiratory instability in the movement of the mandible. The REM stage is usually entered without cortical activation that can be captured by the EEG and without head movement. Therefore, the accelerometer will not measure anything, while the gyroscope will observe changes in the rotation of the mandible. This shows the importance of having both signals from the gyroscope and from the accelerometer in order to correctly observe the entry into the REM stage. The exit from the REM stage is usually closely related to brain activation, which will be observed by the accelerometer and gyroscope, which will observe the isolated jaw movement (IMM) and, where applicable, the movement of the head. As a first approach, the level of the normalized mean value can be considered. For example, changes in amplitude and frequency will also be looked for. Detecting REM during the first 15 minutes of sleep enables the diagnosis of narcolepsy.

Example 7

In a seventh example, see FIG. 14 .

A comparative analysis of the entire sleep, such as the changes in the values of the amplitude and/or frequency of the movements of the mandible and/or other statistical features of the signal, individually or grouped into arrays targeted by a classifier of the random forest type, is applied to practice statistical inference, able to distinguish the different stages. To this end, FIG. 14 shows a spectrogram of the distribution of the frequency of mandibular movements for distinguishing the stages. In this FIG. 14 , the vertical axis represents the amplitude density and the horizontal axis represents the frequency. These specific features of each sleep stage can also be identified by machine deep learning. This algorithm and/or statistical method can also be used to characterize respiratory events and non-respiratory movement events.

The following table gives an example of the variance of the amplitude levels of the mandibular rotation signal between various sleep stages. In this table, "Interval" refers to the interval between the upper layer at the 2.5th percentile and the lower layer at the 97.5th percentile. "Amplitude" refers to the difference between the maximum and the minimum value. "Variance" is a measure of the spread of the values considered. The measurements are based on 1000 samples acquired over a period of 30 seconds for each stage.

stage Variance (mm ² ) Interval(mm) Amplitude(mm) N1 0.016 to 0.064 0.378 to 0.945 0.470 to 1.070 N2 0.018 to 0.052 0.465 to 0.795 0.560 to 0.990 N3 0.081 to 0.143 0.899 to 1.245 1.020 to 1.330 REM 0.004 to 0.098 0.250 to 1.370 0.320 to 1.580

The analysis unit is suitable for verifying whether the standardized mean and variance of the amplitude and frequency of the received first time stream and the second time stream have a level characterizing the REM sleep state during the second time period, in particular during the 30-second time period, and if the standardized mean and variance of the amplitude and frequency of the received first time stream and the second time stream have a level characterizing the REM sleep state, the analysis unit is suitable for generating a third data item representing the REM sleep state.

The identification unit is adapted to identify motion signals characterizing the rotation of the mandible and/or the head movement of the subject in the first time stream and the second time stream. The analysis unit is adapted to analyze these signals, for example by applying these motion signals to a bandpass filter and an exponential moving average or a measurement of the entropy of the frequency of the signal. By applying the bandpass filter, for example to the respiratory frequency, and applying the exponential moving average, for example with a half-life equal to 5, 60, 120 or 180 seconds, to the supplied first and second signal streams and having a first observation period of 30 seconds, the analysis unit will be able to observe whether the signals are unstable. If this is the case, an awakening situation will be observed. On the other hand, if the signal is stable, a sleeping situation can be observed.

The analysis unit is adapted to apply the exponential moving average over a second time period of the first time stream and the second time stream between 30 seconds and 15 minutes, in particular 3 minutes, as a profile characterizing a sleep state. For some analyses, the second time period may even be 30 minutes. The analysis unit is adapted to verify whether the exponential moving average has a substantially constant value during the second time period, and to generate a first data item representing a sleep state or a wake state if the value is substantially constant or not.

The identification unit is adapted to identify motion signals characterizing the rotation of the mandible and head of the subject in the first time stream and the second time stream. The analysis unit is adapted to calculate entropy on the frequencies of these motion signals. By applying this entropy function with an analysis window of, for example, 90 seconds to the supplied first and second signal streams and with an observation period of 30 seconds, the analysis unit can then observe the level of the standardized mean value. If the level is high, an N1 or REM sleep condition will be observed depending on the value of the level.

The identification unit is adapted to identify motion signals characterizing the rotation of the mandible and/or the movement of the head of the subject in the first time stream and the second time stream. The analysis unit is adapted to apply a bandpass filter or a lowpass filter to these motion signals. By applying a bandpass filter, for example at the respiratory frequency, or a lowpass filter (e.g. below 0.10 Hz) to the supplied first and second signal streams and having an observation period of 30 seconds, the analysis unit will be able to observe the level of the standardized mean and/or median. Depending on the level, an N2 or N3 sleep condition will be observed.

Brain activation in the form of micro-awakening has a duration between 3 seconds and 15 seconds (including 3 seconds and 15 seconds) and can be of cortical or subcortical type. Brain activation that causes awakening lasts for more than 15 seconds. Cortical brain activation in REM sleep can be performed by repeating the lowering characteristics of the mandible. In the case of cortical activation, the corticobulbar reflex is activated and multiple sudden movements of large amplitude or even long duration of the mandible are observed. This reflex amplifies the movement. In the case of subcortical activation, this reflex is not activated, and then it is possible to observe only one sudden movement of smaller amplitude in the case of frequency discontinuity relative to the breathing frequency at which the mandible is activated. This movement can have a much lower amplitude and a shorter duration than when the corticobulbar reflex is activated. Therefore, the movement is usually less significant, and the movement can be assisted by detecting the accompanying movement of the head that may only be applied over a very short distance.

Example 8

In another example, reference is made to Table 2. Table 2 shows cortical and subcortical brain activation characteristics. The result of the cortical activation will be a sudden closing or opening of the mandible with large amplitudes for a duration between 3 seconds and 15 seconds. If such cortical activation occurs during sleep, it will typically be accompanied by a change in the position of the subject's head. The analysis unit will use the first data stream and the second data stream to analyze the amplitude and duration of the movement over a 10 second window.

Subcortical activation is characterized by the varying frequency of the mandibular movement and the discontinuity of its shape. The mandible usually remains stable. The analysis unit will use the first and second data streams to analyze the amplitude and duration of the movement over a 10 second window. This analysis can be performed on continuous variables as well.

Therefore, the analysis unit will verify whether the amplitudes of the signals of the received first and second time streams have a level characterizing cortical or subcortical activity during a third time period, in particular a time period between 3 seconds and 15 seconds. If the amplitudes of the received first and second time streams have a level characterizing cortical or subcortical activity, the analysis unit is adapted to generate a sixth data item representing cortical or subcortical activity.

In order to detect the presence of respiratory events or non-respiratory movement events, the analysis unit will analyze the evolution of the position of the mandible, the amplitude of the peak-to-peak jaw movement, the changes in the peak-to-peak amplitude of the jaw movement representing changes in the brain control of the amplitude and the frequency of the jaw movement. If low amplitudes are observed, that is to say amplitudes corresponding to those observed during quiet breathing movements, and in the presence of a stable centrality (jaw movement occurs around a continuous and stable degree of opening of the mouth), no event of sleep disturbance is considered.

If high respiratory control amplitudes are observed (eg, amplitudes corresponding to movements exceeding 0.3 mm), that is, amplitude changes greater than those observed during quiet movement, increased movement or respiratory effort is inferred that may indicate a sleep disturbance.

A central respiratory event is inferred if a large respiratory control amplitude reduction is observed, that is, an amplitude as low as, for example, about 0.1 mm or zero, with stable or unstable control centrality, lasting for at least 10 seconds or, for example, two respiratory cycles.

Measuring the gain of the muscle response of the upper airway during an event enables determination of its obstructive character, i.e. significant respiratory effort, as opposed to no respiratory effort or with reduced respiratory effort, ultimately its central character below a level considered normal. This analysis enables characterization of apneas and hypopneas as obstructive or central. For each sleep stage, the normal level of respiratory effort is predetermined during normal breathing periods during sleep.

The change in the position of the head can be modified with or without the configuration of the event of a change in the sleep stage or the configuration of the transition between sleep and wakefulness. The gain of the muscle reaction during the event is calculated by measuring the peak-to-peak amplitude change during the phasic jaw movement at the respiratory frequency during the event. During the phasic movement of the event, the measurement of the peak-to-peak amplitude difference between the beginning and the end of the event period can already be calculated from a single breathing cycle, the peak-to-peak amplitude difference supplies the gain value. The change can be minimal, of the order of 1/10 mm, or even smaller, but can reach 3 cm. This change can be accompanied by a change in the absolute position of the mandible, meaning that the mouth is more or less open when its phasic movement is applied. Considering the position of the head during sleep, the change can occur in any direction between horizontal and vertical.

Example 9

In a ninth example, reference is made to Table 3. Table 3 shows a typical behavior of brain control for detecting respiratory events and non-respiratory movement events. It can be seen that in order to detect obstructive apnea-hypopnea, the analysis unit will, for example, use the median and/or mean values over the first time stream and the second time stream. An observation time of at least two respiratory cycles or 10 seconds will be preferred in order to make the analysis more reliable. Obstructive apnea-hypopnea is characterized by large brain control amplitudes at respiratory rates that can be repeated cyclically or non-cyclically. It will end with large jaw movements during brain activation. In particular, the distribution of the amplitude values of the jaw movements in the considered stream will be analyzed.

For detecting respiratory effort related to arousal (RERA), the analysis unit will proceed in the same manner as described in the previous paragraph. For detecting central apnea-hypopnea, the duration of observation will also be at least two breathing cycles or 10 seconds.

Example 10

In a tenth example, see FIG. 4 and FIG. 7 .

The presence of teeth grinding will be detected by using, for example, the median, mean, maximum or other statistical values of the rotation of the mandible and its acceleration during an observation time of, for example, 30 seconds.

Brain control following brain activation with or without changes in head position may:

Stable and with low amplitude;

Increase with a high or "rising" centrality; the mouth closes and the event is corrected if the latter is obstructive;

Increases with low or "falling" centrality; mouth is open and events are imposed, obstructive;

·Decreasing with low or “falling” centrality; events are imposed, central;

Increase with a high or "rising" center when recruiting the lateral pterygoid muscles.

If the position of the head does not change but the level of respiratory control changes, then the position of the mandible and the changes therein continue to provide information about the level of respiratory control. FIG7 shows a signal representing cortical brain activation, plotting the movement of the mandible as measured by the accelerometer and gyroscope. In the direction from left to right in this FIG7 , first several oscillations are seen representing the mandible moving at a regular frequency. This movement of the mandible is caused by breathing with a certain degree of effort. The subject involved must make an effort to get air through the upper airway, which can be seen in the amplitude of the signal from the gyroscope. In particular, reference numeral 1 indicates a micro-arousal caused by cortical activation. Then strong oscillations representing movements of larger amplitudes following from brain activation are seen. It can then be seen that the level of the signal has increased, indicating that the mouth has closed and the mandible has risen by a few tenths of a millimeter. If the mandible rises and the respiratory frequency at which its movement is still detected has an amplitude greater than normal, it can be inferred that there is a persistent obstructive event as observed here. If the movement becomes low amplitude, it can be stated that the respiratory control is no longer rising above normal and the respiratory effort has been normalized. Reference numeral 2 indicates micro-arousals caused by subcortical activation, which produce signals with lower amplitude than cortical activation.

The result set obtained after the analysis unit has processed the signal can be presented, for example, in the following manner:

Hypnogram: records the evolution of sleep stages and wake/sleep transition moments during the recording;

Record start and end times, time spent in bed and/or lying down;

Total sleep time; various efficacy indexes;

Sleep fragmentation, such as the number and index of micro-arousals and arousals (activations), wakefulness/

The number and index of sleep transition changes;

·The number and index of respiratory and non-respiratory motion events;

For example, in the case of repetitive central respiratory events with cyclical, periodic, increasing-decreasing variance of brain-controlled amplitude, if the duration of the cycle is measured to be greater than 40 seconds, one might suspect that the type of periodic breathing is evolving, possibly in the context of cardiac insufficiency;

An event of a cyclic nature may also be a blocking type event, or (for example, when the loop gain is high and/or the arousability is strong) a blocking type event may also repeat in a cyclic manner;

Repetitive subcortical activation in response to isolated respiratory effort suggests its association with limb movement.

The position of the head affects the frequency and nature of respiratory and non-respiratory movement events that occur during sleep. Changes in the position of the head during sleep always occur simultaneously with excitations from the brain. During the latter, the head will find a new position, and in the case of such changes, the mandible, which has already moved with a large amplitude through repeated movements, will find a new position that will be subjected to respiratory drive again thereafter, and the amplitude of the respiratory drive will be a measure of the central control level. Therefore, there is a correlation of events, as shown in Figure 7. Therefore, active respiratory control of central activation, possible modifications of the position of the head and the possible modifications of the position of the mandible accompanying it and the modification of the amplitude of the respiratory movement of the mandible is integrated in the brain. The relationship between activity and brain activation is examined. If the control activity level changes, such as the peak-to-peak amplitude of the change in the respiratory movement of the mandible, this is first of all the result of the change in the activation and control state of the brainstem center. In addition, the respiratory activity is captured by the gyroscope during the rotational movement of the mandible, while the central activation is captured by the accelerometer during the linear movement of the head.

Whether the event is a respiratory event or a non-respiratory motion event, changes in the position of the head and the brain activation that accompanies it determine the risk of the event occurring by modifying the level of control and therefore the type of event. Changes in the position of the head are necessarily accompanied by brain activation and can modify the flow conditions of air-fluid in the airways, particularly by modifying the caliber muscle retention conditions of the upper airways, and furthermore the new orientation assumed by the head can expose those airways to mechanical compressive forces.

Mandibular movement and repositioning during cortical or subcortical activation can be described during events as follows:

(1) The mandible is passive, or it drops when activation subsides without the tension and/or phasic support of the musculature, while central motor control is sidelined for the duration of several respiratory cycles. After the mouth is closed and with the mouth passively open, there is no relaxation of its position, that is, no support for the mandible, because the loss of the tension component of the musculature that is considered to support the mandible exceeds a variable distance but with a marked slope (>1/10 mm/s); the measurement of this distance between the lowest point recorded before the closure of the mouth and the change in slope that follows is a marker of passive collapse of the pharynx when following activation, with the presence of a loss of central control; this situation can last for a time equivalent to several (up to five) respiratory cycles.

This relaxation may not occur and the mouth remains closed or virtually closed because there is no loss of central control (persistence of the tonic component) of the musculature that controls the position of the mandible.

(2) The mandible then shows a muscle reaction gain in the form of phases and/or tension, which will reposition it at the respiratory control frequency during this event before a new activation is triggered. This results in a recovery of the muscle activity that controls the position and movement of the mandible. This restored activity of the muscle can be represented by a change in the slope that describes its new position with or without respiratory movement (that is to say with or without a phase component), that is, where at least one peak-to-peak amplitude can be measured beyond the background noise of the measurement (>0.05mm). The latter movement represents a recovery of the respiratory movement, that is to say a shift in the respiratory frequency, and therefore, according to the evaluation rules, the respiratory effort will make it possible to define the obstructive event as central or mixed. The movement of this centrality makes it possible to specify whether the degree of mouth opening is stable, increasing or decreasing, while the peak-to-peak amplitude of the respiratory movement reflects the current degree of effort.

(3) The centrality or amplitude point reaches a minimum and from this centrality or amplitude point a movement to close the mouth is executed, the first movement determined by the activation being similar to the arousability threshold. This movement is sometimes downward, for example in REM or when the mouth is not yet open, because the respiratory effort is first exerted by the activity of the lateral pterygoid muscles and the masseter muscles that hold the mouth in a forward and high position, and this movement bears witness to the activation. The latter can be cortical or subcortical or have a subcortical and then cortical sequence, or when the mandible has not opened much during the event, as if due to the activity of the lateral pterygoid muscles, it can then suddenly open during the activation, while most often, when the mouth is already open during the event, the activation suddenly closes it.

(4) As shown in Figure 4, the mandibular position point at the maximum distance from the arousable point during activation is followed. The distance separating them is a measure of the amplitude of the mandibular movement during activation. This value is measured before activation via the change in amplitude of the respiratory movement from the beginning of the recovery of the effort until the arousable point and compared to the level of respiratory effort deployed during the event. The ratio of these values is a measure of the degree of mandibular loop gain.

Example 11

In an eleventh example, see Figure 8. Figure 8 shows an example of a first time stream F1 (measured by an accelerometer) and a second time stream F2 (measured by a gyroscope) in the case of a subject suffering from obstructive apnea. In this figure, F5n represents the nasal flow and F5th represents the oronasal thermal flow. It will be observed here that during the time period from T1 to T2 and after the apnea represented by reference numeral 1, the signal is unstable. At the beginning of this time period, it is seen that the signal supplied by the gyroscope has a lower amplitude than at the end of the event. Central control is enhanced during this event because it is necessary to combat the obstacle that causes the apnea or hypopnea. During the same time period T1-T2, it can be seen that the accelerometer (reference numeral 2) and the gyroscope (reference numeral 3) indicate respiratory effort, followed by brain activation (reference numeral 4).

Analysis of the signal shows that in the presence of an obstructive apnea between T1 and T2, a movement of the mandible at the respiratory frequency is observed on the accelerometer (F1) and the amplitude increases from the peak-to-peak amplitude, while the mouth opens due to the increasingly decreasing position of the mandible from one respiratory cycle to another (A). At the same time (C), it can be seen that the rotation of the increased amplitude, the angular velocity of the respiratory movement indicates that the effort itself increases. Note the effect of the activation on the movement of the mandible measured with the accelerometer at the level of the letter B, which causes an upward movement of the mandible and has the result of closing the mouth. In this case, the mandible will assume a new position. At the level of the letter D on the gyroscope, it can be seen that this movement of closing the mouth is not purely rotational. The change in the signal state at the level of the letters B and D is simultaneous with the resumption of ventilation in the case of brain activation (micro-arousal).

Example 12

In the twelfth example, see Figure 9. Figure 9 shows an example of a first time stream F1 and a second time stream F2 in the case of an obstructive respiratory hypopnea represented by arrow 1. Arrow 0 represents a wakeful state. This same Figure 9 also shows a sixth time stream F6 representing the presence of snoring captured by an audio sensor, together with a seventh time stream F7 sensed by chin electromyography and an eighth time stream F8 sensed by electroencephalography. It is seen that there is a series of jaw movements (R) of greater amplitude over a period of several seconds, which each time represent cortical or subcortical activation. These movements are accompanied by peak changes in the sixth time stream F6, the seventh time stream F7, and the eighth time stream F8. In fact, the electromyography and electroencephalography signals clearly show that there is brain activation in this case. Obstructive respiratory hypopnea is represented by arrows 2 and 3, arrow 2 represents effort and opening of the mouth, and arrow 3 represents effort and rotation of the mandible. This respiratory hypopnea is followed by activation in the form of micro-arousals, represented by arrow 4. The high values of the respiratory jaw movements between micro-arousals reflect the high respiratory efforts that are also enhanced by snoring. Thus, it can be seen in the first time stream F1 from the accelerometer and in the second time stream F2 from the gyroscope that during snoring there is a rotation of the mandible with the opening of the mouth. Thus FIG9 shows that brain activity can be recorded by an accelerometer and a gyroscope which measure the jaw movement during effort at as much the respiratory frequency as during brain activation, but in this case at a frequency which is no longer typically the respiratory frequency. Still in this FIG9 , the number 0 indicates the subject's state of wakefulness.

Example 13

In a 13th example, see Figure 10. Figure 10 shows an example of a first time stream F1 and a second time stream F2 in the case of a subject suffering from mixed apnea. As in Figure 8, in this Figure 10 it can be seen that the angular velocity of the mandible increases at a frequency corresponding to the respiratory rate. Number 1 represents a lack of respiratory flow, which is associated with a lack of control and effort represented by number 2, followed by a restoration of brain control and effort represented by number 3.

Example 14

In the 14th example, see Figure 11. Figure 11 shows an example of a first time stream F1 and a second time stream F2 in the case of a subject suffering from central apnea. The peaks F show the movement of the head and mandible when breathing is restored. It is also seen that between the peaks F, there is no movement of the mandible, so to speak. The number 1 represents the lack of breathing flow, which is associated with the lack of effort represented by the number 2 and the activation and restoration of effort represented by the number 3.

Example 15

In the 15th example, see Figures 12 and 13. Figure 12 shows an example of the first time stream F1 and the second time stream F2 in the case of a subject suffering from a temporary disappearance of all control of brain origin, which is characteristic of central respiratory hypopnea. This disappearance is characterized by the mouth opening passively because it is no longer bound by muscles. It is therefore visible in the first time stream F1 and the second time stream F2 that between the peaks, the signal does not represent any activity. On the other hand, at the moment of the peak, a high amplitude of the movement of the mandible is observed. Towards the end of the peak, a movement corresponding to a non-respiratory frequency is seen, which is the result of brain activation, which will then lead to micro-awakening. Number 1 indicates a period of hypopnea, in which the reduction of the flow is clearly visible on the fifth time stream F5th from the thermistor. Numbers 2 and 3 represent the disappearance of the mandibular movement during central hypopnea in the first time stream F1 and the second time stream F2. Figure 13 shows an example of the first time stream F1 and the second time stream F2 in the case of a subject experiencing a prolonged respiratory effort that will terminate in brain activation. It can be seen that the signal from the accelerometer F1 indicates a large movement of the head and mandible at the position indicated by H. Thereafter, the second time stream F2 remains almost constant, while the level in F1 from the accelerometer decreases, which indicates that there is in any case a movement of the mandible, which slowly decreases. This is then followed by a high peak I, which is the result of the change in position of the head during the activation that terminates the effort period. Number 1 indicates this long effort period marked by snoring. It can be seen that the effort increases over time, as indicated by number 2. This effort ends in a brain activation, as indicated by number 3, which causes a movement of the head and mandible, as indicated by the letter I.

The analysis unit saves in its memory models of these various signals which are the result of processing using artificial intelligence as described above. The analysis unit will process these streams using those results to produce a report on the analysis of those results.

Accelerometers have been found to be particularly suitable for measuring movements of the head, while gyroscopes, which measure rotational movements, have been found to be particularly suitable for measuring rotational movements of the mandible. Thus, brain activations that result in rotation of the mandible without the head changing position can be detected by the gyroscope. On the other hand, IMM-type movements, especially if the head moves in this case, will be detected by the accelerometer. RMM-type movements will be detected by the gyroscope, which is highly sensitive to RMM-type movements.

Example 16

In further examples, reference is made to an exemplary process for feature extraction, data processing, and data description used in the methods and apparatus provided herein. Such a process is schematically illustrated in FIG. 15 .

Specifically, feature extraction, data processing, and description were done in the R statistical programming language, while machine learning experiments were conducted using sci-kit learn and the SHAP package in Python language.

23 different features were extracted from the raw signal of jaw movement for each event or every 10 seconds of normal breathing. These features included: central tendency of MM amplitude (mean, median and mode); MM distribution (raw or envelope signal): skewness, kurtosis, IQR, 25th percentile, 75th percentile and 90th percentile; extreme values: minimum, maximum, 5th percentile and 95th percentile of MM amplitude; trend: linear trend of MM in time function and coefficients of spline factors (S1, 2, 3, 4) based on tensor product from generalized additive model; duration of each event.

The impact of various features on the model's classification into central hypopnea, normal sleep, and obstructive hypopnea can be described with the help of the SHAP score. The SHAP score measures the average marginal contribution across all possible combinations with other features to classify the 3 target labels. The higher the SHAP score, the more important contribution the feature can provide. Lundberg's Shapley Additive Explanations (SHAP) method unifies the Shapley score (1953) in cooperative game theory (Lloyd S Shapley. "A value for n-person games". In: Contributions to the Theory of Games 2.28 (1953), 307-317.) and local explanation methods (Marco Tulio Ribeiro, Sameer Singh, Carlos Guestrin. "Why should I trust you? Explaining the predictions of any classifier". In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM. 2016, 1135-1144.) to provide the best solution to explain any black box model. The SHAP theory treats the input features as "players" in a cooperative game, where the "payout" is the correct prediction of the target label (i.e., central or obstructive hypopnea). The SHAP algorithm combines each feature value with other features in a random order to form a combination, and then assigns a payout (SHAP score) to each feature value based on their contribution to the total prediction. The SHAP score is the result of averaging the change in the prediction obtained by the combination when the new feature participates. In essence, the SHAP score of a feature value is the average marginal contribution of that feature value to a particular prediction across all possible combinations.

Specifically, the features are extracted as follows:

1. Load a sequence of raw MM data at (e.g., sampling rate = 10 Hz or 25 Hz). The sequence has a significant duration, e.g., between 30 minutes and 8 hours;

2. Timestamps for obstructive and central respiratory hypopnea events;

3. For each timestamp ti, perform the following steps:

3.a. Check whether ti represents the onset of an obstructive or central hypopnea event.

3.b. If ti is the beginning of an obstructive or central hypopnea event,

- assign ti to (t_begin), and then search for the end (t_end); and

- extract the original data sequence into a temporary holder named "EventE" by indexing t_begin and t_end;

4. For each event E, perform the following steps:

4.a. Calculate the event duration dt = (t_end – t_begin)

4.b. Determine the distribution of the parameter measured during the event;

- minimum, maximum, mean, median, mode, 5th percentile, 25th percentile, 75th percentile, 90th percentile, 95th percentile, skewness, kurtosis, IQR;

- fitting a GAM nonlinear model to estimate the MM amplitude and/or position by a spline function over time t, and then extracting the coefficients of the spline function;

- Fit simple linear models and extract intercepts and linear slopes;

- All features + labels are concatenated by matching the measured data with the mandibular motion classes.

After feature extraction, the extracted features and the corresponding target labels are integrated into a tabular dataset.

After that, exploratory data visualization, one-way ANOVA, and paired student-t tests with Bonferroni correction were performed to classify jaw movement characteristics into 3 groups: normal breathing, obstructive hypopnea, and central hypopnea. For null hypothesis testing, the significance level was set to a highly stringent standard (p = 0.001) (10).

For the purpose of model development, the data were randomly split into 2 subsets: a large set (70%) for model development and a small set (30%) for model validation. Because the original training set was unbalanced between central (minority class) and obstructive hypopnea (majority class), the Synthetic Minority Over-sampling Technique (SMOTE, the Synthetic Minority Over-sampling Technique, which is well known per se) on the training set before model development was applied.

A multi-class classification rule is built to classify 3 groups using 23 input features. This consists of a random forest algorithm that combines 500 different decision trees (each decision tree is built on a random subset of 5 features).

The content of the random forest model was then analyzed to assess the importance of each feature and the possible joint contributions to the classification (potential combinations between them to distinguish obstructive hypopnea from central hypopnea). To assess the contribution of each feature to the prediction, the Shapley additive explanation (SHAP) method of Lundberg was used, which is itself well known in the art.

These methods allow, inter alia, the detection of obstructive and central hypopnea.

Example 17

In a further example, see Figures 16 and 17. These Figures show the analysis of jaw movement data captured with the aid of a magnetic sensor. The data analysis is similar to the data analysis of jaw movement data captured with the aid of an accelerometer and/or a gyroscope in addition to the magnetic sensor.

Figure 16 shows 18 most important MM signal features derived from magnetometer measurements, which are sorted by their global impact on the prediction of the model. These bars represent the average SHAP score of each feature, which is layered by 3 target labels, central respiration attenuation (dark gray), normal respiration attenuation (light gray) and obstructive respiration attenuation (gray). The SHAP score measures the average marginal contribution across all possible combinations with other features to classify the 3 target labels. The higher the SHAP score, the more important the contribution that the feature can provide.

FIG. 17 shows an explanation of an event based on the extracted features and based on the SHAP score. Specifically, FIG. 17 shows the SHAP score scale and the probability of the target label. FIG. 17 includes two general regions: region a) and region b. Region a) includes extracted features that support the prediction of the target label, and region b) includes extracted features that point away from the target label.

Example 18

In a further example, see FIG18 , which illustrates an exemplary method for determining sleep stages based on jaw movement data captured with the aid of a gyroscope and an accelerometer. The steps discussed below correspond to the reference numerals in FIG18 .

Specifically, the steps are as follows:

(1) Using the system of the present invention including a gyroscope and an accelerometer to record jaw movement during the subject's sleep. The collected data packets include 6-channel raw signals collected by the three-axis accelerometer and gyroscope sensor. The raw data may also include records from other devices suitable for determining sleep stages, such as EEG, EOG and EMG signals for sleep staging, and 6-channel MM signals collected by the three-axis accelerometer and gyroscope sensor.

(2) The raw data will pass through the preprocessing and feature generation module, after which the raw data is continuously segmented into epochs of 30 s length. Preprocessing involves generating time series sampled at 0.1 Hz and 0.034 Hz (30 s sliding window) from the sleep score series and the time series collected by the sensor and PSG. This preprocessing occurs in two steps: the series or sequence is segmented and then the feature extraction function is applied to each window.

Handcrafted feature extraction can be used as input data for machine learning experiments. For example, the feature generation module uses a sliding window centered on each 30-second period to extract a set of 1728 features from 6 channels of the MM activity signal. The extracted features include: signal energy in the low frequency band (0-0.1 Hz), high frequency band (>0.3 Hz), or respiratory band (0.2-0.3 Hz); exponential moving average with several half-life cycles; entropy of energy in several frequency bands; statistical features applied to the above features: centrality (mean, median) trend, extreme values (minimum, maximum), quartiles, standard deviation, and normal standardized values of all the above features.

(3) The extracted feature set will be fed into a machine learning classifier to generate a soft prediction score (i.e., probability) and binary output for each target label according to a specific classification task. The automated sleep segmentation task is approached at three levels of complexity. The task targets the basic three sleep stages: wake, non-REM (including N1, N2, N3) and REM.

Feature selection and hyperparameter tuning were performed using cross-validation, where the input data was randomly split into packets at the participant level. The final model was trained on the entire training set using only the most relevant features and optimized hyperparameter values. Due to the imbalanced ratio between target labels, the training data was balanced by synthetic minority oversampling technique (SMOTE) before each training session.

Machine learning algorithm: Extreme Gradient boosting (XGB) classifier is used as the core algorithm for all three classification tasks. The XGB classifier is optimized during the training process by minimizing a regularized objective function that combines a convex loss function (based on the difference between the predicted output and the target output) and a penalty term for model complexity.

Model Training: The learning objective was set to multi-class classification, aiming to classify 3 target labels according to a specific task. The training implied the Dropout-Multiple Additive Regression Trees (DART) enhancer and the histogram-optimized approximate greedy tree construction algorithm. Logarithmic loss was chosen as the evaluation metric (thus optimizing the balanced accuracy between the 3 target classes). To prevent overfitting, the learning rate (eta or step shrinkage) parameter was set to 0.01, which shrinks the feature weights to make the enhancement process more conservative.

The output of the model implies a soft-max function to generate probability scores for each target label, and then the final decision (assigning only one label to each 30s segment) is achieved by applying the argmax function on these 3 probability scores.

(4) The most satisfactory solution was adopted based on the epoch-by-epoch agreement between the model's predictions and the reference PSG scores on an unseen validation dataset. Further quantitative evaluation was performed to verify whether the selected algorithm can provide reliable estimates of sleep quality scores such as TST, sleep efficiency, REM ratio, etc. Model selection was based on the following criteria:

Class-by-class consistency evaluation: The normalized confusion matrix allows the class-by-class performance of the model to be evaluated for a specific multi-class classification task. The rows are the true values derived from the manual PSG scoring, and the columns represent the results of the automatic algorithm scoring. The diagonal cells of the confusion matrix represent the class-by-class true positive rate.

Precision (or positive predictive value) measures the model's ability to correctly identify positive cases and is defined as true positives/(true positives + false positives); recall (also called sensitivity, hit rate, or true positive rate) represents the model's utility and is defined as the fraction of correctly classified cases among all target instances: recall = true positive predictions/all positive instances;

The F1 score is a combined metric defined as the harmonic mean of per-class recall and precision:

2*(precision*recall)/(precision+recall)

The F1 score has an intuitive meaning, which represents the accuracy of the model (how many epochs it classified correctly), as well as the robustness of the model (low misclassification rate). Since real-life data presents an unbalanced ratio between sleep stages and all labels are equally important, a classifier that gets an equally high F1 score on all classes is used.

Global epoch-by-epoch consistency evaluation metrics: Balanced accuracy (BAC) measures the average of the true positive rate and true negative rate in the target class. Cohen's Kappa coefficient measures the strength of agreement between the model's classification and the true observations (manual PSG scores). It can be interpreted as 6 levels of consistency strength: below 0: poor, 0-2: slight, 0.2-0.4: fair, 0.41-0.6: moderate, 0.61-0.8: basically, 0.81-1: almost perfect.

(5) The predicted data from the selected (3-category task) model will pass through the interpretation module. The first submodule (sleep score calculation) will convert the sequence of predicted sleep stages into quantitative scores.

The definitions of these quantitative scores are presented in the table below:

(6) Hypnogram creation: A custom function converts a sequence of discretely encoded labels (e.g., 2 = wake, 1 = Rem sleep, 0 = non-REM sleep) into a hypnogram. This graph presents a step line representing discrete sleep stage values as a function of time, which simulates a traditional hypnogram obtained from manual PSG scoring.

Example 19

Example 19 presents an experimental continuation of Example 18. Specifically, the method presented in Example 18 was performed on a group of 96 participants, who were randomly assigned to a training subset (n=68, 70%) and a validation subset (n=28, 30%). Both subsets represent a population of healthy adults ranging in age from 18 to 53 years old.

The system of the present invention including a gyroscope and an accelerometer is used to record jaw movements during the subject's sleep. The collected data packets include 6 channels of raw signals collected by the three-axis accelerometer and gyroscope sensors. The original model is used to develop an automated sleep staging model. In addition, reference data is recorded using equipment suitable for determining sleep stages (such as EEG, EOG and EMG). The latter data is used to determine the accuracy of the applied model.

Instead of using a deep learning model, the traditional framework is followed, which means hand-crafted feature extraction and structured data driven algorithms. Compared to black box models like convolutional neural networks, hand-crafted feature extraction allows better control and understanding of the input data. XGBoost is used for classification tasks. This algorithm provides several advantages over classical methods (LDA, SVM, RF) including high efficiency of computation and resources, allowing very fast training and execution speeds.

Subject Subsets: Polysomnography (PSG) profiles from a group of 96 participants indicated normal sleep activity in both subset groups, with median sleep efficiencies of 89.4% and 87.3%. Within each set, the data structure also presented a proportional imbalance between the 3 sleep stages: in addition to regular wake labels in most cases, non-REM sleep dominated over REM sleep in both groups (92.3 vs. 7.7 for the training set, and 79.9 vs. 20.1 for the validation set), suggesting that data balancing techniques are needed during model development, and that performance metrics should be carefully interpreted during model validation.

3-Class Scoring: This model is designed to classify wakefulness (no sleep), non-REM, and REM sleep. The model results in a well-balanced accuracy between the 3 classes (82.9%, 74.9%, and 82.5% for wakefulness, non-REM, and REM sleep, respectively). The model also has a substantially consistent strength (Kappa = 0.71). It performed best in detecting wakefulness periods with an F1 score of 0.86. Guided by the distribution of wakefulness, non-REM, and REM instances in the model, it was found that identifying wakefulness was easier than distinguishing between non-REM and REM, because the wakefulness instances were well clustered and clearly separated from the other instances, while most of the REM labels were more dispersed and mixed into other non-REM or wakefulness points. This pattern suggests that non-linear algorithms such as random forests, XGboost, or deep neural networks can be considered for successful separation of the 3 classes.

Consistency Analysis of Sleep Quality Index: 3-Category Task Sleep Staging Algorithm can automatically classify each 30-second epoch as wake, non-REM, or REM. The output is then transformed by a second algorithm to provide an estimate of the sleep quality index. Those indices can be divided into 3 main categories: a) time-based indices, which measure the cumulative time (in minutes) in sleep (TST) or during a specific sleep stage (such as wakefulness, REM, or non-REM); b) ratio-based indices, which are estimated as the percentage of epochs in a specific sleep stage (REM, non-REM) over all sleep; c) latency-based indices, which measure the elapsed time between the start of the recording and the onset of sleep (sleep latency) or between the onset of sleep and the first REM epoch (REM latency).

The quantitative score of the automated sleep staging algorithm was determined according to the table presented in Example 18. The difference between the standard score of the PSG profile and the quantitative score of the automated sleep staging algorithm is presented in the following table:

The data showed that the 3-category-based scoring algorithm allowed to measure total sleep time with acceptable accuracy (median difference of only -7.15 minutes, 97.5% CI: -20.34 to +4.38) compared with the reference method (manual PSG scoring). The agreement was also good for determining sleep efficiency (median difference: -1.29%; -3.03 to +0.01).

Conclusion: The feasibility of using the system of the present invention comprising a gyroscope and an accelerometer to record jaw movements during sleep of a subject was explored. The results demonstrate that automatic sleep stage detection based on data measured by a gyroscope configured to measure rotational movement provides better performance at all three resolution levels (for 3-category scoring) compared to a prior art system comprising only an accelerometer.

Claims (16)

1. A system for detecting sleep states of a subject having a head and a mandible,

the system includes a sensing unit configured for mounting on a mandible of the subject, a data analysis unit, and a data link;

wherein the sensing unit comprises a gyroscope configured for measuring rotational movement of the mandible of the subject; and an accelerometer configured for measuring acceleration indicative of the movement and/or position of the head and/or mandible of the subject;

the data link is configured to transmit measured rotational motion data from the gyroscope to the data analysis unit and measured acceleration data from the accelerometer to the data analysis unit;

Wherein the data analysis unit comprises a memory unit configured to store N mandibular movement categories, wherein N is an integer greater than 1, wherein at least one of the N mandibular movement categories represents that the subject is awake, wherein a plurality of the N mandibular movement categories represent that the subject is asleep;

wherein each j (1. Ltoreq.j. Ltoreq.N) th mandibular movement category includes a j-th set of rotation values and/or a j-th set of acceleration values, each j-th set of rotation values representing at least one rate, rate change, frequency and/or amplitude of mandibular rotation associated with the j-th category, each j-th set of acceleration values representing at least one mandibular movement and/or head movement associated with the j-th category;

wherein the data analysis unit comprises a sampling element configured to sample the measured rotational motion data and the measured acceleration data during the same sampling period, thereby obtaining sampled rotational motion data and acceleration data;

wherein the data analysis unit is configured to:

deriving a plurality of measured rotational and acceleration values from the sampled rotational and acceleration data; matching the measured rotation and acceleration values with the N mandibular movement categories; and

Detection indicates that the subject is in a sleep state of waking or sleeping.

2. The system of claim 1, further comprising a magnetometer adapted to measure magnetic field data, changes in magnetic field data representing movements of the subject's head and/or mandible,

the data link is further configured for transmitting measured magnetic field data from the magnetometer to the data analysis unit;

wherein each jth (1. Ltoreq.j. Ltoreq.N) mandibular movement category includes a jth set of magnetic field data values, each jth set of magnetic field data values representing at least one rate or change in rate of mandibular movement or head movement associated with the jth category;

wherein the data analysis unit comprises a sampling element configured to sample the measured magnetic field data during a sampling period, thereby obtaining sampled magnetic field data;

wherein the data analysis unit is configured to derive a plurality of measured magnetic field values from the sampled magnetic field data; and

wherein the data analysis unit is further configured for matching the measured magnetic field values with the N mandibular movement categories.

3. The system of any one of claims 1 or 2, wherein at least one of the N mandibular movement categories represents a change in position of the head; wherein the data analysis unit is further configured to identify a movement of the head of the subject based on the rotational movement and acceleration data to discern the movement of the head from a movement of a mandible of the subject.

4. A system according to claim 3, wherein the data analysis unit is configured to:

analyzing an evolution of the position of the mandible, the position of the head, a peak-to-peak amplitude of mandibular movement, a change in peak-to-peak amplitude of mandibular movement, a frequency of mandibular movement, and/or a change in frequency of mandibular movement; and

sleep states are detected when a change in peak-to-peak amplitude of the mandibular motion and/or frequency of the mandibular motion is observed in the evolution of the mandibular motion.

5. The system of claim 4, wherein one or more of the N mandibular movement categories are characterized by a predetermined frequency range that includes frequencies representing respiration of the subject and includes frequencies between 0.15Hz and 0.60 Hz.

6. The system of claim 4, wherein one or more of the N mandibular movement categories are characterized by a predetermined frequency range that includes frequencies representing respiration of the subject and includes frequencies between 0.25Hz and 0.50 Hz.

7. The system of claim 4, wherein one or more of the N mandibular movement categories are characterized by a predetermined frequency range that includes frequencies representing respiration of the subject and includes frequencies between 0.30Hz and 0.40 Hz.

8. The system of any of claims 5-7, wherein at least one of the N mandibular movement categories indicates that the subject is in an N1 sleep state; wherein at least one of the N mandibular movement categories indicates that the subject is in REM sleep; wherein at least one of the N mandibular movement categories indicates that the subject is in an N2 sleep state, and/or wherein at least one of the N mandibular movement categories indicates that the subject is in an N3 sleep state.

9. The system of claim 8, wherein the data analysis unit is configured to detect a respiratory disorder during a sleep state indicative of the subject being asleep; wherein one or more of the N mandibular movement categories represent obstructive apneas, mixed apneas, obstructive hypopneas, wake-related respiratory effort, central apneas and/or central hypopneas.

10. The system of claim 9, wherein the data analysis unit is configured to:

analyzing an evolution of the position of the mandible, the position of the head, a peak-to-peak amplitude of the mandibular movement, a change in the peak-to-peak amplitude of the mandibular movement, a frequency of the mandibular movement and/or a change in the frequency of the mandibular movement; and

the presence of a respiratory disorder event is detected when a change in peak-to-peak amplitude of the mandibular motion and/or frequency of the mandibular motion is observed in the evolution of the mandibular motion.

11. The system of claim 10, wherein one of the N mandibular movement categories represents molar, wherein the measured rotational movement data represents mandibular movement amplitude of at least 1mm at a frequency established in the range of 0.5Hz to 5Hz during at least three respiratory cycles, or mandibular movement amplitude exceeding 1mm for at least 2 seconds in a sustained, stressed manner, when the movement is stepwise.

12. A method for detecting a sleep state of a subject having a head and a mandible, comprising the steps of:

measuring a rotational movement of the mandible of the subject by means of a gyroscope and an acceleration representing the movement and/or position of the head and/or mandible of the subject by means of an accelerometer; wherein the gyroscope and accelerometer are positioned on a mandible of the subject;

Receiving, by a data analysis unit and via a data link, measured rotational motion data from the gyroscope and measured acceleration data from the accelerometer;

storing N mandibular movement categories by means of a memory unit comprised in the data analysis unit, wherein N is an integer greater than 1, wherein at least one of the N mandibular movement categories represents that the subject is awake, wherein a plurality of the N mandibular movement categories represent that the subject is asleep;

wherein each j (1. Ltoreq.j. Ltoreq.N) th mandibular movement category includes a j-th set of rotation values and/or a j-th set of acceleration values, each j-th set of rotation values representing at least one rate, rate change, frequency or amplitude of mandibular rotation associated with the j-th category, each j-th set of acceleration values representing at least one mandibular movement and/or head movement associated with the j-th category;

sampling the rotational movement data and the measured acceleration data during the same sampling period by means of a sampling element comprised in the data analysis unit, thereby obtaining sampled rotational movement data and acceleration data;

Deriving a plurality of measured rotational and acceleration values from the sampled rotational and acceleration data by means of the data analysis unit;

matching the measured rotation and acceleration values with the N mandibular movement categories by means of the data analysis unit; and

detection indicates that the subject is in a sleep state of waking or sleeping.

13. The method of claim 12, further comprising the step of:

measuring magnetic field data by means of a magnetometer, the change of the magnetic field data representing a movement of the head and/or mandible of the subject;

transmitting measured magnetic field data from the magnetometer to the data analysis unit by means of the data link;

wherein each jth (1. Ltoreq.j. Ltoreq.N) mandibular movement category includes a jth set of magnetic field data values, each jth set of magnetic field data values representing at least one rate or change in rate of mandibular movement or head movement associated with the jth category;

sampling the measured magnetic field data during a sampling period by means of a sampling element comprised in the data analysis unit, thereby obtaining sampled magnetic field data;

Deriving a plurality of measured magnetic field values from the sampled magnetic field data by means of the data analysis unit; and

by means of the data analysis unit, the measured magnetic field values are matched to the N mandibular movement categories.

14. The method according to any one of claims 12 or 13, wherein at least one of the N mandibular movement categories represents a change in position of the head; the method further includes identifying a motion of the head of the subject based on the rotational motion and acceleration data to discern the motion of the head from the motion of the mandible of the subject.

15. The method of claim 14, wherein the data analysis unit is configured to:

analyzing an evolution of the position of the mandible, the position of the head, a peak-to-peak amplitude of mandibular movement, a change in peak-to-peak amplitude of mandibular movement, a frequency of mandibular movement, and/or a change in frequency of mandibular movement; and

sleep states are detected when a change in peak-to-peak amplitude of the mandibular motion and/or frequency of the mandibular motion is observed in the evolution of the mandibular motion.

16. The method of claim 15, wherein at least one of the N mandibular movement categories indicates that the subject is in an N1 sleep state; wherein at least one of the N mandibular movement categories indicates that the subject is in an N2 sleep state, wherein at least one of the N mandibular movement categories indicates that the subject is in a REM sleep state; and/or wherein at least one of the N mandibular movement categories indicates that the subject is in an N3 sleep state.